An Evaluation of the Joint Operation of Ramp Metering and Variable Speed Limits Using Corsim

MISSING IMAGE

Material Information

Title:
An Evaluation of the Joint Operation of Ramp Metering and Variable Speed Limits Using Corsim
Physical Description:
1 online resource (226 p.)
Language:
english
Creator:
Mintsis, Evangelos
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Master's ( M.S.)
Degree Grantor:
University of Florida
Degree Disciplines:
Civil Engineering, Civil and Coastal Engineering
Committee Chair:
Elefteriadou, Ageliki L
Committee Members:
Washburn, Scott S
Yin, Yafeng

Subjects

Subjects / Keywords:
corsim -- metering -- vsl
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre:
Civil Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Rampmetering is a traffic management scheme which has been widely implemented andtested over the past 50 years. It has been shown that ramp metering is anefficient, cost-effective and viable method for managing freeway traffic andexploiting freeway capacity. VSL is a traffic control measure which has beenmostly applied around work-zones, school-zones and freeway locations whererecurring congestion occurs. VSL diminish shockwaves upstream of bottlenecks byhomogenizing traffic flow. The combined operation of these two systems has beenexamined with the use of macroscopic simulation so far. Hypothetical networkshave been developed in METANET for the testing of the integrated operation oframp metering and VSL. This study investigates the effects of the concurrentoperation of ramp metering and VSL on traffic flow with the use of microscopicsimulation (i.e. CORSIM). The study site is a 13-mile section of the I-95 inMiami, which has been calibrated and validated according to field datacollected from the STEWARD database. A test-bed is created for the simulationof two VSL algorithms (i.e. occupancy based VSL and volume based VSL) and tworamp metering algorithms (i.e. Fuzzy Logic ramp metering algorithm and ALINEAalgorithm) in CORSIM. The integrated control scenario is tested against the nocontrol case, the VSL only case and the ramp metering only case. The results(i.e. network wide and link statistics) indicate that the integrated controlcase outperforms the rest of the traffic management scenarios.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Evangelos Mintsis.
Thesis:
Thesis (M.S.)--University of Florida, 2012.
Local:
Adviser: Elefteriadou, Ageliki L.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2013-02-28

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044754:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 AN EVALUATION OF THE JOINT OPERATION OF RAMP METERING AND VARIABLE SPEED LIMITS USING CORSIM By EVANGELOS MINTSIS A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 Evangelos Mintsis

PAGE 3

3 To my family

PAGE 4

4 ACKNOWLEDGMENTS I thank my advisor, Dr. Lily Elefteriadou, Professor, Department of Civil and Coastal Engineering, University of Florida, for her guidance and support throughout the duration of my thesis. I would also like to thank my committee members, Dr. Scott Washburn, Associate Professor, and Dr. Yafeng Yin, Associate Professor, for their guidance and feedback on the study. I thank Clark Letter for providing the computer program he developed for the replication of the operation of the variable speed limits (VSL) in CORSIM. I wou ld like to give special thanks to Dr. Angelos Barmpoutis, Assistant Professor, Digital Worlds Institute, University of Florida, for his invaluable assistance in computer programming and Fuzzy Logic. Finally, I would like to thank my family and friends for their constant moral support and encouragement.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ....... 4 LIST OF TABLES ................................ ................................ ................................ .................. 7 LIST OF FIGURES ................................ ................................ ................................ ............... 9 LIST OF ABBREVIATIONS ................................ ................................ ................................ 15 ABSTRACT ................................ ................................ ................................ ......................... 16 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .......... 18 1.1 Background ................................ ................................ ................................ ............ 18 1.2 Problem Statement ................................ ................................ ................................ 1 9 1.3 Research Objectives ................................ ................................ .............................. 20 1.4 Document Organization ................................ ................................ ......................... 20 2 LITERATURE REVIEW ................................ ................................ ............................... 22 2.1 Ramp Metering ................................ ................................ ................................ ...... 22 2.1.1 The Classification of Ramp Metering Systems ................................ .......... 24 2.1.2 Requirements for a Successful Implementation of Ramp Metering .......... 26 2.2 Variable Speed Limits (VSL) ................................ ................................ ................. 26 2.3 Coordination of VSL and Ramp Metering ................................ ............................. 31 2.4 Literature Review Summary ................................ ................................ .................. 36 3 METHODOLOGY ................................ ................................ ................................ ......... 40 3.1 CORSIM ................................ ................................ ................................ ................. 40 3.2 Algorithms Tested ................................ ................................ ................................ .. 41 3.2.1 VSL Algorithm Based on Occupancy Measureme nts ................................ 41 3.2.2 VSL Algorithm Based on Flow Measurements ................................ ........... 42 3.2.3 ALINEA Ramp Metering Algorithm ................................ ............................. 43 3.2.4 Fuzzy Logic Ramp Metering Algorithm ................................ ....................... 44 3.3 Scenarios ................................ ................................ ................................ ............... 46 3.4 Analysis Process ................................ ................................ ................................ .... 48 4 THE SIMULATION NETWORK FOR I 95 IN MIAMI ................................ .................. 60 4.1 The Study Site ................................ ................................ ................................ ........ 60 4.2 Configuration of the Network ................................ ................................ ................. 60 4.2.1 Network Geometry ................................ ................................ ....................... 60

PAGE 6

6 4.2.2 Incoming Volumes ................................ ................................ ....................... 61 4.2.3 Implementation of the Ramp Metering Algorithm ................................ ....... 62 4.2.4 Detector Placement throughout the Network ................................ ............. 62 4.3 Calibration Process ................................ ................................ ................................ 63 5 ANALYSIS OF THE SIMULATION RESULTS ................................ ........................... 88 5.1 Traffic Impacts of the Simulated Traffic Management Schemes ........................ 88 5.1.1 No control Case ................................ ................................ ........................... 89 5.1.2 VSL only Case ................................ ................................ ............................. 92 5.1.3 Ramp Metering only Case ................................ ................................ ........... 97 5.1.4 Integrated Control Case ................................ ................................ ............ 103 5.1.5 Statistical Analysis of the Simulation Results ................................ ........... 108 5.2 Interactions between Ramp Metering and VSL ................................ ................. 109 5.2.1 Low Demand Scenario ................................ ................................ .............. 110 5.2.2 Medium Demand Scenario ................................ ................................ ........ 113 5.2.3 High Demand Scenario ................................ ................................ ............. 115 5.2.4 Summary of the Interactions between Ramp Metering and VSL ............ 119 6 SUMMARY AND CONCLUSIONS ................................ ................................ ............ 215 APPENDIX: CALIBRATION DATA ................................ ................................ .................. 220 LIST OF REFERENCES ................................ ................................ ................................ .. 223 BIOGRAPHICAL SKETCH ................................ ................................ ............................... 226

PAGE 7

7 LIST OF TABLES Table page 3 1 Occupancy thresholds for displayed speed limits (Scenario 1). ........................... 50 3 2 Occupancy thresholds for displayed speed limits (Scenario 2). ........................... 50 3 3 Volume thresholds for displayed speed limits (Sc enario 1). ................................ 50 3 4 Volume thresholds for displayed speed limits (Scenario 2). ................................ 50 3 5 Description of fuzzy logic ramp metering algorithm inputs. ................................ ... 51 3 6 Rule base for fuzzy ramp metering algorithm. ................................ ....................... 51 3 7 Scenarios evaluating traffic on the I 95 in Miami when no traffic management scheme is implemented. ................................ ................................ ......................... 52 3 8 Scenarios to evaluate the operation of the proposed VSL algori thms. ................ 53 3 9 Scenarios evaluating the operation of the ALINEA and the Fuzzy Logic Ramp Metering Algorithms. ................................ ................................ ............................... 54 3 10 Scenarios evaluating the joint operation of the ALINEA ramp metering and the volume based VSL algorithms. ................................ ................................ ......... 55 3 11 Scenarios evaluating the joint operation of the ALINEA ramp metering and the occupancy based VSL algorithms. ................................ ................................ ... 56 3 12 Scenarios evaluating the joint operation of the Fuzzy Logic ramp metering and the volume based VSL algorithms. ................................ ................................ 57 3 13 Scenarios evaluating the joint operation of the Fuzzy Logic ramp metering and the occupancy based VSL algorithms. ................................ ............................ 58 3 14 Modified rule weights of fuzzy ramp metering algorithm rule base. ..................... 59 4 1 Location and metering rate of each ramp signal along I 95 ................................ .. 66 4 2 Location of field and simulated detectors ................................ ............................... 66 4 3 Average network speed for each simulation run ................................ ................... 67 4 4 Field and simulated speeds and volumes 1 st time period ................................ .. 67 4 5 Field and simulated speeds and volumes 2 nd time period ................................ 68 4 6 Field and simulated speeds and volumes 3 rd time period ................................ .. 68

PAGE 8

8 4 7 Field and simulated speeds and volumes 4 th time period ................................ .. 69 4 8 Field and simulated speeds and volumes 5 th time period ................................ .. 69 4 9 Field and simulated speeds and volumes 6 th time period ................................ .. 70 4 10 Field and simulated speeds and volumes 7 th time period ................................ .. 70 4 11 Field and simulated speeds and volumes 8 th time period ................................ .. 71 4 12 Field and simulated speeds and volumes 9 th time period ................................ .. 71 4 13 Field and simulated speeds and volumes 10 th time period ................................ 72 4 14 Field and simulated speeds and volumes 11 th time period ................................ 72 4 15 Field and simulated speeds and volumes 12 th time period ................................ 73 5 1 Cumulative network statistics on the I 95 in Miami under no traffic management scheme. ................................ ................................ ........................... 122 5 2 Cumulative network statistics on the I 95 in Miami during the operation of the tested VSL algorithms. ................................ ................................ .......................... 123 5 3 Cumulative netwo rk statistics on the I 95 in Miami during the operation of the tested Ramp Metering algorithms. ................................ ................................ ........ 124 5 4 Cumulative network statist ics on the I 95 in Miami during the concurrent operation of the ALINEA ramp metering and the volume based VSL algorithm. ................................ ................................ ................................ ............... 125 5 5 Cumulative network statistics on the I 95 in Miami during the concurrent operation of the ALINEA ramp metering and the occupancy based VSL algorithm. ................................ ................................ ................................ ............... 126 5 6 Cumulative network statistics on the I 95 in Miami during the concurrent operation of the Fuzzy Logic ramp metering and the volume based VSL. ........ 127 5 7 Cumulative network statistics on the I 95 in Miami during the concurrent operation of the Fuzzy Logic ramp metering and the occupancy based VSL. ... 128 5 8 Results of the conducted statistical analysis. ................................ ....................... 129 A 1 Entering and exiting percentages in the simulated network. ............................... 221 A 2 Car following sensitivity parameter by link. ................................ .......................... 222

PAGE 9

9 LIST OF FIGURES Figure page 2 1 Active ramp meters at a two lane motorway in Munich Germany. ...................... 39 4 1 The I 95 Expressway at NW 62nd Street before the modifications ...................... 74 4 2 The I 95 Expressway at NW 62nd Street after the modifications ......................... 74 4 3 Location of detectors used for calibration on I 95 in Miami. ................................ .. 75 4 4 Field and simulated speeds 1st time period ................................ ....................... 76 4 5 Field and simulated volumes 1st time period ................................ ...................... 76 4 6 Field and simulated speeds 2nd time period ................................ ...................... 77 4 7 Field and simulated volumes 2nd time period ................................ .................... 77 4 8 Field and simulated speeds 3rd time period ................................ ....................... 78 4 9 Field and simulated volumes 3rd time period ................................ ..................... 78 4 10 Field and simulated speeds 4th time period ................................ ....................... 79 4 11 Field and simulated volumes 4th time period ................................ ..................... 79 4 12 Field and simulated speeds 5th time period ................................ ....................... 80 4 13 Field and simulated volumes 5th time period ................................ ..................... 80 4 14 Field and simulated speeds 6th time period ................................ ....................... 81 4 15 Field and simulated volumes 6th time period ................................ ..................... 81 4 16 Field and simulated speeds 7th time period ................................ ....................... 82 4 17 Field and simulated volumes 7th time period ................................ ..................... 82 4 18 Field and simulated speeds 8th time period ................................ ....................... 83 4 19 Field and simulated volumes 8th time period ................................ ..................... 83 4 20 Field and simulated speeds 9th time period ................................ ....................... 84 4 21 Field and simulated volumes 9th time period ................................ ..................... 84 4 22 Field and simulated speeds 10th time period ................................ ..................... 85

PAGE 10

10 4 23 Field and simulated volumes 10th time period ................................ ................... 85 4 24 Field and simulated speeds 11th time period ................................ ..................... 86 4 25 Field and simulated volumes 11th time period ................................ ................... 86 4 26 Field and simulated speeds 12th time period ................................ ..................... 87 4 27 Field and simulated volumes 12th time period ................................ ................... 87 5 1 Average speed per link under four different traf fic control schemes for the low demand scenario between 16:15 16:30 pm. ................................ ..................... 130 5 2 Average speed per link under four different traffic control schemes for the low demand scenario between 16:45 17:00 pm. ................................ ..................... 131 5 3 Average speed per link under four different traffic control schemes for the low demand scenario between 17:15 17:30 pm. ................................ ..................... 132 5 4 Average speed per link under four different traffic control schemes for the low demand scenario between 17:45 18:00 pm. ................................ ..................... 133 5 5 Average speed per link under four different traffic control schemes for the low demand scenario between 18:15 18:30 pm. ................................ ..................... 134 5 6 Throughput north of Opa Locka Blvd. under all the examined traffic management schemes for the low demand scenario. ................................ ......... 1 35 5 7 Throughput north of Biscayne Canal under all the examined traffic management schemes for the low demand scenario. ................................ ......... 136 5 8 Travel time per vehicle from I 195 EB/WB till the exit t o 95th St. under all the examined traffic management schemes for the low demand scenario. ............. 137 5 9 Travel time per vehicle from NW 81st St. till the exit to North Golden Glades Dr. under all the examined traffic management schemes for the low demand scenario. ................................ ................................ ................................ ................. 138 5 10 Delay per vehicle on the on ramp from NW 125th St. under all the examined traffic management schemes for the low demand scenario. ............................... 139 5 11 Delay per vehicle on the on ramp from Opa Locka Blvd. under all the examined traffic management schemes for the low demand scenario. ............. 140 5 12 Average speed per link under four different traffic control schemes for the medium demand scenario between 16:15 16:30 pm. ................................ ...... 141 5 13 Average speed per link under four different traffic control schemes for the medium demand scenario between 16:45 17:00 pm. ................................ ...... 142

PAGE 11

11 5 14 Average speed per link under four different traffic control schemes for the medium demand scenario between 17:15 17:30 pm. ................................ ...... 143 5 15 Average speed per link under four different traffic control schemes for the medium demand scenario between 17:45 18:00 pm. ................................ ...... 144 5 16 Average speed per link under four different traffic control schemes for the medium demand scenario between 18:15 18:30 pm. ................................ ...... 145 5 17 Throughput north of Opa Locka Blvd. under all the examined traffic management schemes for the medium demand scenario. ................................ 146 5 18 Throughput north of Biscayne Canal under all the examined traffic management schemes for the medium demand scenario. ................................ 147 5 19 Throughput south of US 441 for the best implementation of each traffic management scheme during the medium demand scenario. ............................. 148 5 20 Travel time per vehicle from I 195 EB/WB till the exit to 95th St. under all the examined traffic management schemes for the medium demand scenario ...... 149 5 21 Travel time per vehicle from NW 81st St. till the exit to North Golden Glades Dr. under all the examined traffic manage ment schemes for the medium demand scenario. ................................ ................................ ................................ .. 150 5 22 Delay per vehicle on the on ramp from NW 125th St. under all the examined traffic management schemes for the medium demand scenario. ....................... 151 5 23 Delay per vehicle on the on ramp from Opa Locka Blvd. under a ll the examined traffic management schemes for the medium demand scenario. ...... 152 5 24 Average speed per link under four different t raffic control schemes for the high demand scenario between 16:15 16:30 pm. ................................ ............ 153 5 25 Average speed per link under four different traffic control schemes for the high demand scenario between 16:45 17:00 pm. ................................ ............ 154 5 26 Average speed per link under four different traffic control schemes for the high demand scenario between 17:15 17:30 pm. ................................ ............ 155 5 27 Average speed per link under fo ur different traffic control schemes for the high demand scenario between 17:45 18:00 pm. ................................ ............ 156 5 28 Average speed per link under four different traffic control schemes for the high demand scenario between 18:15 18:30 pm. ................................ ............ 157 5 29 Throughput north of Opa Loc ka Blvd. under all the examined traffic management schemes for the high demand scenario. ................................ ....... 158

PAGE 12

12 5 30 Throughput north of Biscayne Canal under all the examined traffic management schemes for the high demand scenario. ................................ ....... 159 5 31 Travel time per vehicle from I 195 EB/WB till the exit to 95th St. under all the examined traffic management schemes for the high demand scenario. ............ 160 5 32 Travel ti me per vehicle from NW 81st St. till the exit to North Golden Glades Dr. under all the examined traffic management schemes for the high demand scenario. ................................ ................................ ................................ ................. 161 5 33 Delay per vehicle on the on ramp from NW 125th St. under all the examined traffic management schemes for the high demand scenario. ............................. 162 5 34 Delay per vehicle on the on ramp from Opa Locka Blvd. under all the examined traffic management schemes for the high demand scenario. ............ 163 5 35 One VSL sign located just downstream of the entrance from Opa Locka Boulevard. ................................ ................................ ................................ .............. 164 5 36 One VSL sign located immediately upstream of the entrance from Opa Locka Boulevard. ................................ ................................ ................................ .............. 165 5 37 Installation of two VSL signs on the northbound direction of the I 95 in Miami. 166 5 38 Interactions between ramp metering and VSL during Scenario 16 A. ............... 167 5 39 Interactions between ramp metering and VSL during Scenario 16 B. ............... 168 5 40 Interactions between ramp metering and VSL during Scenario 16 D. ............... 169 5 41 Interactions between ramp metering and VSL during Scenario 16 E. ............... 170 5 42 Interactions between ramp metering and VSL during Scenario 19 A. ............... 171 5 43 Interactions between ramp metering and VSL during Scenario 19 B. ............... 172 5 44 Interactions between ramp metering and VSL during Scenario 19 D. ............... 173 5 45 Interactions between ramp metering and VSL during Scenario 19 E. ............... 174 5 46 Interactions between ramp metering and VSL during Scenario 10 A. ............... 175 5 47 Interactions between ramp metering and VSL during Scenario 10 B. ............... 176 5 48 Interactions between ramp metering and VSL during Scenario 10 D. ............... 177 5 49 Interacti ons between ramp metering and VSL during Scenario 10 E. ............... 178 5 50 Interactions between ramp metering and VSL during Scenario 13 A. ............... 179

PAGE 13

13 5 51 Interactions between ramp metering and VSL during Scenario 13 B. ............... 180 5 52 Interactions between ramp metering and VSL during Scenario 13 D. ............... 181 5 53 Interactions between ramp metering and VSL during Scenario 13 E. ............... 182 5 54 Interactions between ramp metering and VSL during Scenario 17 E. ............... 183 5 55 Interactions between ramp metering and VSL during Scenario 17 B. ............... 184 5 56 Interactions between ramp metering and VSL during Scenario 17 D. ............... 185 5 57 Interactions between ramp metering and VSL during Scenario 17 E. ............... 186 5 58 Interactions between ramp metering and VSL during Scenario 20 A. ............... 187 5 59 Interactions betwee n ramp metering and VSL during Scenario 20 B. ............... 188 5 60 Interactions between ramp metering and VSL during Scenario 20 D. ............... 189 5 61 Interactions between ramp metering and VSL during Scenario 20 E. ............... 190 5 62 Interactions between ramp metering and VSL during Scenario 11 A. ............... 191 5 63 Interactions between ramp metering and VSL during Scenario 11 B. ............... 192 5 64 Interactions between ramp metering and VSL during Scenario 11 D. ............... 193 5 65 Interacti ons between ramp metering and VSL during Scenario 11 E. ............... 194 5 66 Interactions between ramp metering and VSL during Scenario 14 A. ............... 195 6 67 Interactions between ramp metering and VSL during Scenario 14 B. ............... 196 5 68 Interactions between ramp metering and VSL during Scenario 14 D. ............... 197 5 69 Interactions between ramp metering and VSL during Scenario 14 E. ............... 198 5 70 Interactions between ramp metering and VSL during Scenario 18 A. ............... 199 5 71 Interactions between ramp metering and VSL during Scenario 18 B. ............... 200 5 72 Interactions between ramp metering and VSL during Scenario 18 D. ............... 201 5 73 Interactions between ramp metering and VSL during Scenario 18 E. ............... 202 5 74 Interactions between ramp metering and VSL during Scenario 21 A. ............... 203 5 75 Interactions between ramp metering and VSL during Scenario 21 B. ............... 204

PAGE 14

14 5 76 Interactions between ramp metering and VSL during Scenario 21 D. ............... 205 5 77 Interacti ons between ramp metering and VSL during Scenario 21 E. ............... 206 5 78 Interactions between ramp metering and VSL during Scenario 12 A. ............... 207 5 79 Interactions between ramp metering and VSL during Scenario 12 B. ............... 208 5 80 Interactions between ramp metering and VSL during Scenario 12 D. ............... 209 5 81 Interactions between ramp metering and VSL during Scenario 12 E. ............... 210 5 82 Interactions between ramp metering and VSL during Scenario 15 A. ............... 211 5 83 Interactions between ramp metering and VSL during Scenario 15 B. ............... 212 5 84 Interactions between ramp metering and VSL during Scenario 15 D. ............... 213 5 85 Interactions between ramp metering and VSL during Scenario 15 E. ............... 214

PAGE 15

15 LIST OF ABBREVIATION S CORSIM Corridor Simulation FDOT Florida Department of Transportation FLC Fuzzy Logic Control FMOH Freeway Management and Operations Handbook O D Origin Destination RMCH Ramp Management and Control Handbook RTE Run Time Extension VMS Variable Message Sign VSL Variable Speed Limits

PAGE 16

16 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science AN EVALUATION OF THE JOINT OPERATION OF RAMP METERING AND VARIABLE SPEED LIMITS USING CORSIM By Evangelos Mintsis August 2012 Chair: Lily Elefteriadou Major: Civil Engineering Ramp metering is a traffic management scheme which has been widely implemented and tested over the past 50 years. It has been shown that ramp metering is an efficient, cost effective and viable method for managing freeway traffic and exploiting freeway cap acity. VSL is a traffic control measure which has been mostly applied around work zones, school zones and freeway locations where recurring congestion occurs. VSL diminish shockwaves upstream of bottlenecks by homogenizing traffic flow. The combined operat ion of these two systems has been examined with the use of macroscopic simulation so far. Hypothetical networks have been developed in METANET for the testing of the integrated operation of ramp metering and VSL. This study investigates the effects of the concurrent operation of ramp metering and VSL on traffic flow with the use of microscopic simulation (i.e. CORSIM). The study site is a 13 mile section of the I 95 in Miami, which has been calibrated and validated according to field data collected from the STEWARD database. A test bed is created for the simulation of two VSL algorithms (i.e. occupancy based VSL and volume based VSL) and two ramp metering algorithms (i.e. Fuzzy Logic ramp metering algorithm and ALINEA algorithm) in CORSIM. The integrated con trol scenario is tested against the no

PAGE 17

17 control case, the VSL only case and the ramp metering only case. The results (i.e. network wide and link statistics) indicate that the integrated control case outperforms the rest of the traffic management scenarios.

PAGE 18

18 CHAPTER 1 INTRODUCTION 1.1 Background Ramp metering was first pressway in 1963 to control the rate at which vehicles entered the facility from on ramps with the use of traffic signals (Ramp Management and Control Handbook RMCH, 2006). Subsequent successful ramp metering experiments that were conducted in Detroit and Los Angeles sh owed that this ramp management strategy is an efficient, viable, and practical method to regulate freeway traffic. Since, numerous ramp meters have been installed at several metropolitan areas in the US and around the world, improving traffic conditions an d mollifying safety and environment al issues. Throughout the years r esearch pertaining to ramp metering has flourished and led to the development of more robust systems which are cap able to sustain a high level of performance at complex and congested freeway facilities. Variable speed limits (VSL) have been also deployed over the last 30 years as a control measure to manage freeway traffic in Europe, Australia and the United States (Freeway Management and Operations Handbook FMOH, 2003). In Germany a total of more than 500 miles of VSL equipped freeway stretches are currently in operation (Papageorgiou et al., 2008). VSL have been designed to dynamically update posted speed limits to accurately assist drivers on speed selection when traffic and enviro nmental conditions are less than ideal (Sisiopiku, 2001). Research and field evaluations have demonstrated that VSL installations can provide a significant safety benefit through the stabilization and homogenization effect that they exert on traffic flow.

PAGE 19

19 It is reported that accident numbers have been reduced by as much as 20 30% after VSL implementation (Papageorgiou et al., 2008). 1.2 Problem Statement Ramp metering is capable of increasing mainline throughput and sustaining traffic flow efficiency when t raffic starts to become dense. However, during excessively high demand periods, congestion occurs even when more sophisticated ramp metering algorithms are implemented. It has been suggested that w hen traffic becomes too dense, VSL can complement ramp mete ring by limiting the inflow at breakdown locations, thus postponing or eliminating the onset of congestion (Hegyi et al., 2005). no study conducted to assess the operations under a combined field implementatio n of ramp metering and VSL. So far, the effects of the joint operation of ramp metering and VSL on traffic flow behavior have been investigated on a theoretical basis with the use of macroscopic simulation and specifically METANET. METANET is a macro simu lator developed to reproduce actual traffic conditions and evaluate a variety of freeway management strategies. Generally, macro simulators describe traffic flow in an aggregate manner, and thus they cannot imitate the individual vehicle movements as dicta ted by the car following, lane changing and gap acceptance behavior of drivers. This study will evaluate the joint operation of ramp metering and VSL at a microscopic level, using the CORSIM (i.e. CORridor microscopic SIMulation) micro simulation program CORSIM is a micro simulator which has been widely used for traffic operations analysis

PAGE 20

20 1.3 Research Objectives The primary objective of this study is to evaluate the impacts of the combined operation of ramp metering and variable speed limits on traffic flow and freeway network performance with the use of microscopic simulation. The following tasks were conducted to achieve these objectives : Completed a literature review on the development, implementation, and operation of the ramp metering and VSL syst ems, as well as on the effects of their integrated control on freeway traffic performance Simulated and calibrated the I 95 Miami corridor in CORSIM. Built a Run Time Extension (RTE) to replicate the Fuzzy Logic ramp metering algorithm and combined it wit h two VSL RTEs (Letter, 2011) to construct the desired set of simulated scenarios. Interface d the RTEs with the Miami I 95 CORSIM network to run the simulation experiments. Evaluate d the simulation results for different combinations of algorithms to invest igate the impacts of the combined control on freeway traffic flow and network performance. Provide d recommendations for the efficient implementation of both systems on a freeway facility. 1.4 Document Organization T he thesis document is organized as follows : Chapter 2 present s a literature review on the introduction, development, field implementation and assessment of the ramp metering and VSL systems, as well as on the studies that have been conducted so far to test the potential of integrating them for congestion mitigation. Chapter 3 will describe the process of testing the combined operation of ramp metering and VSL. Chapter 4 will describe the simulation and calibration of the I 95 Miami freeway network. Chapter 5 will describe the relationships o bserved between the operation of the ramp metering and the VSL algorithms and will present the results of the simulated

PAGE 21

21 scenarios. The last chapter will provide conclu sion s and recommendations regarding the joint operation of the two systems for the diffe rent combination s of algorithms.

PAGE 22

22 CHAPTER 2 LITERATURE R EVIEW 2. 1 Ramp Metering Ramp metering has been developed, as increasing demand for mobility has imposed the necessity for an efficient and cost effective management, operation and maintenance of freeway facilities (RMCH, 2006). Since its debut in the early 1960s, ramp metering has been widely applied in several urban areas worldwide. Ramp metering is defined as number of vehicles entering a freeway in order to achieve operational objectives (Figure 2 1) ( FMOH, 2003). T his technique allows for more consistent flow operations by smoothing the flow of freeway traffic and exploiting more efficiently the existing ca pacity. Previous research has shown that the entrance of platoons of vehicles through unmetered on ramps destabilizes the mainline traffic flow, causing frequently breakdowns (Elefteriadou et al. 1995). ff for a six week study in Minneapolis, Minnesota, a before and after evaluation concluded that meters were responsible for a 21 percent reduction in crashes and a 9 percent increase in mainline A more recent analysis conducted based on data coll ected before and after the shutdown experiment in the Twin Cities, suggest s that ramp metering significantly increases capacity at bottleneck locations (Zhang and Levinson, 2010). Zhang and Levinson (2010) statistically tested several hypothes e s based on a large empirical multi bottleneck dataset in an effort to shed light into the fundamental relationship between

PAGE 23

23 ramp metering and freeway capacity. The results of their test s revealed that ramp metering increases freeway capacity in three manners: It postpo nes the activation of bottlenecks, or completely eliminates them by prolonging the pre queue transition period The average flow rates during the pre queue transition period are higher with the implementation of ramp metering Even after the activation of a bottleneck, ramp metering sustains higher average discharge queue flow rate compared to the unmetered case. The findings of their research exhibit that, unlike conventional beliefs, ramp metering not only contributes to the prevention of breakdowns at b ottlenecks, but it is also increasing throughput even after the breakdown has occurred. Generally, according to the experience gained so far, the following advantages are reported from the deployment of ramp metering as a regulatory traffic strategy: Incre ase in vehicle throughput Increase in vehicle speeds Increase of freeway capacity Increase in consistency (i.e. reliability) of travel time s Reduction in the total number of crashes and crash rate (especially rear end and sideswipe collisions) Reduction of vehicle emissions and fuel consumption Ability to provid e High Occupancy Vehicles (HOV) priority Traffic diversion to under utilized local streets However, ramp metering has also shown to have a number of drawbacks : Equity: Many argue that ramp metering favors suburban motorists that make longer trips versus those who make shorter trips and live within the metered zones.

PAGE 24

24 Socio economic considerations: Ramp metering may shift traffic congestion and associated impacts from one location to another. Adverse impact on surface network: R amp meters create queues on the ramp, which may spill back to the surrounding network. 2.1 .1 The Classification o f Ramp Metering Systems Generally, ramp metering systems can be characterized as (RMCH, 2006) : Pre timed Control T he metering rate is fixed and usually is determined using historic volume data. Meters are activated only on pre set schedules. The control interval over which the selected metering rate is in effect can range from 30 minutes to the entire peak period Thi s type of control is not effective in cases of unexpected events such as incidents and changes in the demand Pre timed control can be applied to either local or area wide levels Local Traffic Responsive Control It is based on detection near the ramp (i mmediately upstream or downstream of the ramp or at the merge point ) Speed, volume, and/or occupancy are typically monitored and the metering rate is set based on these measurements Two types of algorithms belong to this categor y. Demand/Capacity algorit hms and occupancy based algorithms. In the demand/capacity algorithms, the metering rates are the difference between the upstrea m flow (measured f r om the detectors) and the downstream capacity. For occupancy algorithms, the metering rates are based on the occupancy measurements upstream (and/or downstream ) of the ramp. T he metering rate is usually estimated and updated every minute. Area wide (or S ystem wide) Traffic Responsive Control It is a coordinated control system that uses speed, occupancy and/or volume data from the field to set the metering rates, on a system wide basis. The purpose of the area wide control is to

PAGE 25

25 optimize traffic flow along a stretch of roadway, rather than at a specific location. Therefore, the metering rates at the ramps are interrelated. The data used in area wide traffic responsive systems are from local ramp and freeway detectors, but also from downstream and/or upstream detectors. This category of systems entails the most complex hardware configuration compared to th e previous metering approaches. Generally, there are three types of area wide control algorithms (Zhang et al., 2001): Cooperative A lgorithms : After computing the metering rate for each on ramp, further adjustment is performed to avoid congestion at the bottleneck and spillback at critical ramps. Examples of such cooperative algorithms are the Helper Algorithm and the Linked Ramp Algorithm. Competitive A lgorithms : In the competitive algorithms, t wo sets of metering rates are computed based on both local and global traffic conditions, and the most restrictive rate is selected to be implemented. Further adjustment to the selected metering rates may also be made to account for spillback and other con straints. Integral Algorithms : Integral ramp metering algorithms have a clear control objective(s) that is explicitly or implicitly linked to the control action. The objective is usually to minimize travel time, or maximize throughput of the entire system. They de termine ramp metering rates through optimizing the objective function while considering system constraints, such as maximum allowable ramp queue, bottleneck capacity, and so forth. This class of algorithms is appealing because of their solid theore tical foundation and their capability of handling various types of metering and modeling constraints. However, these algorithms are also invariably more complex in logic and more demanding in computation. More recent research has proposed the enhancement o f several existing ramp metering algorithms (i.e. COMPA SS, Stratified Zone Metering, Bottleneck, Demand Capacity and ALINEA), so that they consider the probability of breakdown at a merge area and then select the metering rate as a function of a tol erable probability of breakdown (Elefteriadou et al., 2009). The methodology is based on breakdown probability models for all the critical ramps of a freeway site. These models provide the probability of breakdown as a function of the freeway and ramp flows. The

PAGE 26

26 breakdown, and determining the ramp metering rate such that the des ired threshold is not exceeded. The advantage of this approach is that it incorporates the randomnes s of the maximum throughput into the ramp metering rate selection. 2.1 .2 Requirements for a Successful Implementation of Ramp Metering The RMCH (2006) emphasizes that a successful implementation of ramp metering strikes a balance between freeway mainline improvements and vehicle delays and queues on the entrance ramps. Generally the following should be considered: Ramp metering strategy: Selected to reflect the goals and objectives for the specific system (e.g. reduce congestion on the mainline, crash redu ction, etc.), as well as whether the system would be local or system wide, and pre timed or traffic responsive. Geographic extent: The area to be covered by ramp metering and whether the meters will be isolated (i.e. local control) or part of a larger syst em of meters (i.e. coordinated). Ramp metering approaches: The type of algorithm to be used, i.e. local, system wide, pre timed, traffic responsive, etc. Queue management: Consideration of the metering algorithm impacts on ramp queue lengths and potential spillback to the adjacent network. Flow control: The rate at which traffic will be released from the meter; one or two at a time, in one lane or multiple lanes. 2.2 Variable Speed Limits (VSL) The introduction of VSL as a traffic management scheme was syn chronous to that of ramp metering. In the early 1960s the VSL were initially proposed and implemented in order to indicate safe speeds to drivers during not ideal driving conditions (i.e. non recurring and recurring congestion) Since then the applicatio n of VSL has expanded worldwide (i.e. U.S.A., Europe, and Australia ). VSL have been mostly used to increase safety at school zones and work zones (Hines, 2002). Kwon et al. (2007) developed an advisory VSL system for work zones, in

PAGE 27

27 order to decrease mean speeds upstream of the work zone to the same level of speed the traffic flow sustained downstream of it. The system was implemented with the use of three VSL signs (one upstream of the work zone and two downstream) for three weeks at one of the I 494 work zones in Twin Cities, Minnesota. The VSL were set based on speeds collected every 30 sec from several detector stations and varied every 1 minute in a 5 mph increment. The evaluation of the system showed a significant reduction of the average 1 minute maxi mum speed difference along the work zone during the morning peak period, thus creating a more stable flow and increasing safety. However, VSL have been also installed at other freeway sites to provide safety benefits and stabilize the traffic flow. Speed l imits were adjusted in England in response to the level of congestion on the M25 motorway in 1995. The system was designed to smooth traffic flow by reducing stop and go traffic The 22.6 km long syst em had VSL stations spaced at 1 km intervals, loop detectors at 500 meter intervals, and CCTV. Using loop detectors that measur ed traffic density and speed, speed limits were lowered in de crements as congestion was increas ing The sp eed limits were lowered from 70 mph to 60 mph when the volume exceeded 165 0 veh/h/l a n e and lowered to 50 mph when the volume exceeded 2050 veh/h/l a n e Results showed that traffic accidents decreased by 10 15% and there was a very high compliance with the VSL system (Robinson, 2000). Throughout the years several algorithms contr olling the operation of VSL have been developed. Some of them have been implemented at selected facilities, while others have only been tested in simulated environments. Below is a complete list of the algorithms that have been proposed so far:

PAGE 28

28 Algorithms based on flow measurements, which reduce the speed limits as flow thresholds are exceeded (e.g. implemented in M25 motorway in England Robinson, 2000 ). Algorithms based on occupancy measurements, which reduce the speed limits as occupancy thresholds are exceeded Such an algorithm has been implemented in I 4 freeway in Orlando, Florida (Haas et al., 2009) Algorithms based on average travel speeds, which set the speed limits according to the range of speeds the measurement belongs (implemented in PARAMICS simulation and proposed by Lee et al. 2004 ). Algorithms based on logic trees that utilize flow, speed and occupancy measurements (implemented in PARAMICS simulation proposed by Allaby et al. 2007). Algorithms based on the prevailing weather conditions (i.e. visibility, wind speed, precipitation severity, etc.). Such an algorithm has been implemented in the A16 motorway near Breda, in the Netherlands. T he operation of the currently implemented VSL alg orithms is likely not optimal, as it is mostly based on predetermined thresholds of speed, flow or occupancy for decision making. Therefore, our understanding of the actual impact of VSL on traffic flow efficiency remains confined, and there is no solid gr ound for developing appropriate control strategies. As it has been already stressed t he primary effect of VSL on traffic operations throughout a roadway network is the increase of traffic safety. In Europe, VSL have been established at several freeway str etches, where high frequency of car crashes had been reported, and proved to significantly contribute to the ir reduction. Several years of evaluations show a decrease of accidents between 20% and 30% after VSL implementation (Papageorgiou et al., 2008) Ho wever, the impact of VSL on traffic flow efficiency is still not entirely clear The effort to evaluate freeway performance before and after the installation of VSL on selected freeway stretches has not provided a clear answer so far.

PAGE 29

29 Zackor ( 1972, 1991) w as the first one to evaluate the effects of VSL on traffic flow. His analysis was based on traffic data collected from a two lane highway in Germany and consisted of counts both before and after the establishment of the VSL signs. The operation of the VSL system was based on flow measurements and weather conditions (i.e. precipitation, visibility) and the displayed VSL were informative. Zackor concluded that speed differences among individual vehicles and mean accelerations/decelerations diminished in the p resence of VSL. He also indicated that with respect to the fundamental traffic flow diagram at lower and medium levels of traffic volume, the mean speed is lower when VSL are active, while at higher volumes the opposite occurs due to the homogenization eff ect Thus, both speed and capacity rise about 5% to 10% at the same time according to Zackor. titative model by Cremer (1979). model that predict ed the changes that VSL imposed on the shape of the fundamental diagram. According to this model VSL could shift the critical flow at higher values causing active bottlenecks to attain larger capacity. Cremer speculated that lower speed variati ons result in a more stable flow and a more stable flow allows for higher speeds at the same density, thus achieving higher traffic flow. This prediction is in conflict with the results of later research conducted by Smulders (1990 1992 ). Smulders collect ed traffic data (i.e. 1 min interval mean speeds and volumes) during a six weeks experiment at a two lane stretch of freeway near Utrecht and a three lane stretch near Rotterdam. The dataset included counts both with and without the VSL operating. The anal ysis showed a significant reduction of shock

PAGE 30

30 waves after the implementation of VSL. The percent of time headways smaller than 1 sec at the median lane diminished drastically leading to a more stable traffic stream. Smulders could not identify any correlat ion between the implementation of VSL and capacity increase phenomena. Moreover, he stated that mean speeds, lane distribution speed differences between lanes were hardly influenced. However he also mentioned that these results could not be conclusive du e to the limited duration of the experiment and the fact that the VSL were not enforced and were set manually. Hegyi (2004) assumed that the activation of VSL, when density approaches its critical value, simply shifts the current traffic state to a new one of reduced flow and slightly higher density. He indicated that t he decrease of flow would prevent the occurrence of a breakdown, and the VSL would thus contribute to the traffic flow efficiency in this manner. Thus, he suggest s that the portion of the fun damental diagram representing the uncongested regime should be substituted with a straight line with slope corresponding to the displayed VSL value. A more recent and thorough analysis on the topic has been conducted by Papagerogiou, Kosmatopoulos and Pap amichail (2008). The analysis has been based on traffic data collected from a European highway, where a flow/speed threshold based VSL control algorithm is in place. The research team utilized the data to create flow occupancy diagrams corresponding to two different time periods. During the first period the activation of the VSL w as rather sparse due to the initially selected threshold values; this is considered as the non VSL period. During that period the VSL were advisory. During the second period the activation of the VSL w as frequent as the system had

PAGE 31

31 been significantly tuned and the threshold v alues had been modified; t his period is considered as the VSL period. During that period the VSL were enforced. Their results indicate that during under critic al conditions the activation of VSL signs decreases the mean speed of vehicles causing travel times to increase. However, they mention that this effect may also result in a delay of breakdown, as traffic flow was found to decrease. A more interesting obser vation is that VSL seem to move the fundamental diagram to higher critical occupancy (i.e. density) values. Although the analysis suggested that the critical density increases when VSL are active, no such conclusion could be derived for the critical flow. For a number of locations there was indeed a slight visible increase of flow, but for several others no increase was observable Thus, this effort did not provide definite answers regarding a potential capacity increase for VSL installation s Letter (2011) examined the effects of VSL on traffic flow with the use of micro simulation. He integrated three VSL algorithms (i.e. flow based, occupancy based, and one based on flow, speed and occupancy measurements) into CORSIM utilizing the concept of RTEs and test ed them on a stretch of the I 95 freeway in Miami, Florida His results indicate that VSL increased average speeds and decreased travel times over the test network. However, he reported that the network wide benefits were only marginal compared to the posi tive effect of VSL upstream of identified bottleneck locations. At these locations, average speeds increased significantly and queue lengths were reduced substantially compared to the no control case according to his analysis. 2.3 Coordination of VSL and Ramp Metering As it has been already mentioned, r amp metering has been widely applied as a traffic management scheme to mitigate traffic congestion in merging areas. Although

PAGE 32

32 ramp metering has been proven to significantly improve traffic flow operations, breakdown becomes inevitable when traffic demand increases excessively (Hegyi et al. 2005). This phenomenon occurs despite the implementation of more sophisticated metering algorithms, mainly due t o limited ramp storage and equity issues. Several simulation studies that have been condu cted (Papamichail et al., 2008, Ghods et al., 2009, Carlson et al. 2010 Hegyi et al., 2005), indicate that VSL can complement ramp metering and postpone or prevent t raffic breakdown. These studies have been primarily conducted with a macroscopic traffic simulation model called METANET. METANET is a time and space discrete second order model, which has been developed by Messmer and Papageorgiou (1990) in the form of p artial differential equations. It accommodates traffic in an aggregate manner, estimating mean speed, density and flow for stretches of roadways with uniform characteristics. As it is a second order model it is able to represent mean speed dynamics fairly realistically. METANET can simulate various traffic condition s (both congested and uncongested) as well as capacity reducing events (i.e. incidents). Any freeway network can be buil t within the model irrespective of its topology and geometric characteristi cs. Several traffic management schemes, such as ramp metering, variable speed limits, route guidance and shoulder lane opening or closing can be implemented within the model. METANET provides output in the form of macroscopic traffic characteristics. Traff ic density, volume and speed are estimated every time step for every segment of the freeway, as well as travel times on predetermined routes of the network. The results a re aggregated and presented in 1 min output intervals. Additionally, the model

PAGE 33

33 calcula t es global performance measures over the simulation time line, which can be exploited for the assessment of different control strategies. In METANET a network is represented by a graph of directed links. The model includes five different types of links (i .e. normal freeway links, origin links, store and forward links, destination links, and dummy links) and each one is used according to the s to a freeway stretch of uniform (i.e. no on/off ramps) and h omogeneous (i.e. same number of lanes, grade and curvature) characteristics. Nodes are placed at locations where on/off ramps exist or major geometrical changes occur. The traffic states on the five types of links and the nodes are determined by a set of t raffic flow models for each time step. Ramp metering had been included in METANET from the very early version of the model. The ALINEA local ramp metering was the first strategy to be incorporated. In order to embed this control method into METANET, a slig ht modification of the origin link model was proposed. Whenever ramp metering is active, the maximum outflow that leaves an origin during a time period is restricted by a metering rate where is a minimum admissible value. Two different approaches have been developed for the integration of VSL into METANET. The first one has been proposed by Hegyi et al. (2005) and the second one by Papamichail et al. (2008). Both approaches are based on the c orresponding models that the research teams have developed to explain the impacts of VSL on aggregate traffic flow behavior. In both cases the modification of the pre existing normal freeway link model is required, in order for the integration to become a ttainable. Then a specific VSL can be applied at every freeway link separately.

PAGE 34

34 The model developed by Papamichail et al. (2008) assumes that VSL values are assigned in the form of rates which range between and have a lower admissible bound The VSL rates are introduced into the link model by converting the equation which estimates mean speed for each link segment into a dependent function. On the other hand, Hegyi et al. (2005) have proposed a more simplistic approach. They believe that a more realistic effect of the speed limits is achievable, if the desired speed is the minimum of the following two quantities: the desired speed based on the exper ienced density and, the desired speed caused by the speed limit displayed on the variable message sign (VMS) All the studies conducted so far (Papamichail et al., 2008, Ghods et al., 2009, Carlson et al. 2010 Hegyi et al., 2005), have introduced an objec tive function to the represents the total time spent by the vehicles in the network throughout the simulation time line. T his objective function is minimized with the utilization of an appropriate algorithm. The optimization problem is solved for different traffic control strategies (i.e. no control, only ramp metering, coordination of ramp metering and VSL ) and the total time spent is minimized and compared among the various scenarios. In this manner the effect of the combined ramp metering and variable speed limit s performance is examined and compared to that of other traffic control measures. Hegyi et al. (200 4 ) applied a model predictive control framework to optimize the coordination of VSL and ramp metering He constructed a simplified network in

PAGE 35

35 METANET to assess the performance of this framework. The network was loaded with a hypothetical demand scenario, wh ich remained constant across the examination of all the implemented traffic management schemes. The case of coordinated ramp metering and variable speed limits was tested against a ramp metering coordinated only case and a no control case. The total time s pent was reduced by 9.0% during the integrated control compared to the ramp metering only case and 14.3% compared to the no control one. The study concluded that the network wide performance improved substantially after the combined implementation of VSL a nd ramp metering. An optimal control framework has been utilized by two other studies (Papamichail et al., 2008, Carlson et al., 2010), to solve the same optimization problem. In both studies, simplified networks of similar geometry and configuration were developed in METANET, so that different traffic control measures can be tested with the proposed framework. Hypothetical demand scenarios were used in both cases. The integrated control (i.e. combination of ramp metering and variable speed limits) was test ed against no control, only ramp metering and only VSL schemes. At all instances, the integrated control outperforms the other traffic control measures and provides superior performance to the network. Similar results regarding the performance of coordinat ed ramp metering and VSL have been also reported by Ghods et al. (2009) In this study, a fuzzy logic control framework was developed to solve the optimization problem. The controller was fine tuned with the use of a genetic algorithm. The researche r s constructed a simplified network in METANET similar to that of Hegyi et al. (2005) and used the model proposed for the replication of the VSL operation proposed by the latter study. Integrated control

PAGE 36

36 was tes ted against the no control case, the ALINEA ra mp metering algorithm and a coordinated fuzzy ramp metering algorithm The operation of the integrated control proved to be the most beneficial for the performance of the network, as total time spent was decreased significantly compared to the other scenar ios. 2.4 Literature Review Summary Ramp metering is a control measure which regulates traffic entering a freeway facility from on ramps. By restricting the number of vehicles entering the freeway, it reduces density at merge areas thus preventing or delay ing the occurrence of breakdowns. Moreover, it facilitates merging operations by breaking platoons of incoming vehicles, which could cause excessive turbulence at the merging points. The effect of VSL on traffic flow is not as clearly documented. Some res earchers indicate that VSL can efficiently stabilize the traffic stream at the stretches they are applied by suppressing the formation of shock waves that would have otherwise occurred. Other research has reported that there is no improvement in the speeds or maximum throughput when VSL is implemented (Smulders, 1990) Most recent research seems to indicate that the benefits of VSL can be found in the areas upstream of the bottleneck, and that VSL might also lead to a slight overall decrease in travel time through the ext ended freeway section (Letter, 2010) However, with respect to the research that has been conducted so far to investigate the impacts of VSL on traffic flow there are two important issues. Firstly, there is lack of information pertaining to the compliance of drivers to the corresponding VSL systems. Moreover, this information is always presented in a qualitative manner rather in a quantitative and more coherent one. This fact has serious implications regarding the credibility of the analyses, as it has been proven that compliance of

PAGE 37

37 drivers is one of the most significant factors for an efficient implementation of VSL. Secondly, little is mentioned relevant to the location of the VSL signs and the location of the traffic count detectors with re spect to the location of active bottlenecks. A more precise presentation of the sites, the location of the equipment and the configuration of the facilities at these locations would facilitate the understanding of the impact of VSL on traffic flow. The combined implementation of these two systems might result in more efficient operations during periods when demand is excessively high and each one of them is unable to individually alleviate congestion Under these circumstances ramp metering and VSL may be able to either sustain higher flows or minimize travel times compared to an uncontrolled case or when only one of them is operating T raffic simulators are useful, cost effective and convenient tools, which are used for the planning of new and exp ansion of existing transportation facilities. Moreover, they offer the capability of examining the operation of alternative management schemes and their effects on vehicular traffic flow. The examination of the impacts of the coordination of ramp metering and variable speed limits on traffic flow efficiency has been merely done with the deployment of macroscopic traffic flow modeling evaluating the combined operation of these two control methods ba se d on field data Thus, a nalyses have been conducted only on an abstract basis with the use of METANET a macroscopic simulation model METANET is a purely deterministic model as several of its variables describing the aggregate behavior of traffic flow (i.e. free flow speed, critical density) can only

PAGE 38

38 accept fixed predetermined values R andomness has not been incorporated into the tochastic. Additionally there is no consensus pertinent to the impacts of VSL on aggregate traffic flow behavior. Thus, there is doubt regarding the reliability of the existing VSL models incorporated into METANET to fully and accurately replicate the eff ects of VSL on traffic flow It is also essentially critical to be mentioned that, little validation effort has been undertaken to ensure that this modeling approach is capable of replicating accurately real traffic operations (Papageorgiou et al. 2010) Most of the networks that have been tested are simplified ones, constructed by experts, and the demand scenarios that they have been loaded with are hypothetic al Considering the empirical nature of the model, more scholastic validation is necessary based on real traffic counts collected from existing facilities. Therefore, the evaluation of the ramp metering together with VSL in a real world calibrated freeway network would provide additional information regarding their combined impacts on traffic flow. M icroscopic simulators are capable of imitating the actual movement processes of pertaining to the proper simulation of both ramp metering and VSL strategies In the case of ramp metering micro simulation is better able to replicate the gap selection and acceptanc e process and its randomness. In the case of VSL micro simulation would facilitate the examination of the homogenization effect that VSL is supposed to provide.

PAGE 39

39 Figure 2 1. Active ramp meters at a two lane motorway in Munich, Germany.

PAGE 40

40 CHAPTER 3 METHODOLOGY 3.1 CORSIM The assessment of the integrated operation of ramp metering and VSL was selected to be conducted with the use of a microscopic simulation program named CORSIM I 95 in Miami was chosen as the study area and the traffic conditions on this freeway network were replicated within CORSIM. CORSIM is capable of imitating the longitudinal and lateral movemen t of individual vehicles as they occur in real based on a set of sub car following, lane changing and gap acceptance behavior. The progra m has been widely used in North America for traffic operations analyses for several years and its ability to accurately reproduce real life traffic conditions has been well validated. A few ramp metering algorithms have been already incorporated in to CORSI M (i.e. demand/capacity, ALINEA, etc.). T his study will focus on the examination of the ALINEA and the Fuzzy Logic Ramp Metering a lgorithm s The Fuzzy Logic Ramp Metering Algorithm has been implemented along the I 95 corridor in Miami since March 2010 and is not yet included in CORSIM Thus a R TE was built and interfaced with CORSIM to replicate the operation of the latter algorithm within the micro simulator. A CORSIM RTE is an external application which can be directly interfaced with the CORSIM simulati on and override or supplement Although the ALINEA algorithm has been embedded into CORSIM, the program is unable to provid e ramp metering rates as an output. This shortcoming of CORSIM has been surpassed

PAGE 41

41 with the development of an other RTE which can replicate the ALINEA algorithm and export the estimated ramp metering rates as an output concurrently. With respect to VSL, Letter (2011) developed RTEs for incorporat ing three different algorithms into CORSIM This study e valuates two of these algorithms in conjunction with the two abovementioned ramp metering ones. T he first VSL algorithm is currently active on the I 4 in Orlando (PBS&J, 2009) and its operation is determined according to occupancy measurements collected fr om the field. The second one has been implemented on the M25 motorway in England (Robinson, 2000) and its operation is based on detected flow measurements. 3.2 Algorithms Tested A detailed description of the operation of the two ramp metering algorithms (i .e. Fuzzy Logic ramp metering algorithm, ALINEA algorithm) and the two VSL algorithms (i.e. occupancy based algorithm, flow based algorithm) is presented in this section. 3.2 .1 VSL Algorithm B ased on Occupancy Measurements In this case the speed limits are adjusted when occupancy thresholds are exceeded. Each VSL sign is connected with a specific detector station, and the average occupancy counted across all lanes indicates whether the traffic conditions are uncongested, lightly or heavily congested based o n the se predetermined occupancy thresholds. Each traf fic state is characterized by two distinct upper and lower occupancy value s; one pertinent to the increase of the speed limits and one pertinent to the decrease. Accordingly the posted speed limit s are i ncremented or decremented by 5 mph. Thus, the adjusted speed limit never deviate s more than 10 mph from the initially se t value and the adjustment is smooth every update interval. This algorithm is currently operating along the I 4 in Orlando (PBS&J, 2009) In the I 4 implementation of this

PAGE 42

42 algorithm the speed limit s are updated every 120 seconds. I n CORSIM the update interval has been reduced to 60 seconds due to simulation constraints. Two different occupancy thresholds scenarios have been selected for evaluation within this study. In the first scenario the thresholds are similar to those of the I 4 VSL system. The thresholds for the second scenario have been selected according to the findings of the NCHRP Report 3 87 (Elef teriadou et al. 2009). The occu pancy thresholds corresponding to each scenario are presented in Tables 3 1 and 3 2 respectively 3.2 .2 VSL Algorithm B ased on Flow Measurements This algorithm adjusts the speed limits when specific flow thresholds are crossed. Similarly to the occupancy b ased VSL algorithm, each VSL sign is linked to a detector station which calculates the average hourly volume across all lanes and the speed limits are set relative to this measurement. A speed limit may decrease when a specific volume threshold is exceeded but it will revert to the initial value when a different threshold is crossed. This strategy prevents the rapid oscillations of the speed limits. This algorithm has been implemented on the M25 motorway in England (Robinson, 2000). The first set of volume thresholds considered for simulation in this study has been obtained from this implementation of the algorithm and are presented in Table 3 3. A second set of thresholds has been chosen according to speed flow diagrams that have been developed in the 2000 Highway Capacity Manual (HCM, 2000), and are shown in Table 3 4. The thresholds have been determined by identifying the volume of traffic where speed s drop for a given free flow speed, utilizing the associated volume as the threshold point.

PAGE 43

43 3.2 .3 ALINEA R amp Metering Algorithm ALINEA is a local feedback ramp metering algorithm which has been developed based on robust automatic control methods (Smaragdis & Papageorgiou, 2003). This algorithm has been implemented at several sites in many European countries ( e.g. A10 West motorway in Amsterdam, Boulevard P riph to control easily and efficiently traffic entering freeways (Papageorgiou et al., 1997). (e.g. ev ery minute) according to the subsequent formula: r(k) = r(k 1) + K R o out (k)] (3 1) where K R > 0 is a regulator parameter. Fi e l d applications of the algorithm have shown that a value of K R = 1.17 veh/min produces very good results (Papageorgiou et al., 1997) If the computed metering rate lies outside a pre specified range of values [r min r max ], then it is adjusted to remain always within that range. The value of r max is equal to orithm requires the estimation of occupancy from a detector station located downstream of the ramp at a point where the effects of the ramp flow on the breakdown of the mainstream traffic are visible. ALINEA is gradually adjusting the metering rates, so t he traffic flow is finally stabilized around the critical occupancy. Thus the capacity of the freeway merging s egment is efficiently utilized. In this study the critical occupancy has been selected equal to 22%, the m inimum metering rate equal to 5 v eh/min the ma ximum metering rate equal to 28 veh/min and the in itial metering rate equal to 10 veh/min

PAGE 44

44 3.2 .4 Fuzzy Logic Ramp Metering Algorithm This section describes the structure and the operation of the fuzzy l ogic ramp metering algorithm. This algorithm was developed by Meldrum and Taylor (2000) and was implemented by the Washington State Department of Transportation in the Northwest Region of Washington State to manage on line the operation of over 100 on ramps The membership of an element into a set c annot be simply characterized by a element can belong partially to one or more sets. Fuzzy logic control (FLC) utilizes natural linguistic variables, heuristics of human rea soning and rule based logic to process imprec ise or incomplete information. This attribute of FLC is one of the most important that makes it appropriate to ramp metering However, there are a few more reasons why FLC is suitable for ramp metering (Meldrum and Taylor, 2000) : It can utilize incomplete or inaccurate data. It can balance conflicting objectives. It does not require extensive system modeling. It is easy to tune. FLC does not req uire extensive system modeling. The operation of many exis ting algorithms lies on the presumption that freeway capacity is constant at each specific location. Yet it has been shown that capacity changes drastically according to weather conditions, incidents, wor k zones or demand fluctuation. Since FLC is using c ongestion indicators for the estimation of the metering rating it considers implicitly all these factors that af Thus, there is no need for adjustment of the control parameters on a local level. Moreover, the fact that FLC u ses linguistic variables and rule based logic that replicates the way an operator thinks about ramp

PAGE 45

45 metering, allows for the easy tuning of the algorithm, as the performance objective (e.g., longer or shorter queues) are different among various locations. Fuzzy logic control is comprised of three core steps : 1) fuzzification to convert the quantitative inputs into natural language variables; 2) rule evaluation to implement the control heuristics; and 3) defuzzification to transform the qualitative rule outc omes to a numerical output. Fuzzification is t he first step in the FLC D uring this step the inputs are preprocessed to the controller. The inputs to the controller are presented in Table 3 5 Fuzzification corresponds each numerical input into a set of fu zzy classes, also known as linguistic variables. The membership of each input into every fuzzy class is measured on a scale of 0 to 1 indicat ing how true that class is. After fuzzification, the rule base is evaluated. The rules are a set of if then statem ents similar to the heuristics an operator would use to control the system (Table 3 6 ). The rule outcome is equal to the degree of activation of the rule premise. A weighting factor corresponds to e ach rule and determines its relative significance within the rule base. Through the modification of these rule weights, different performance objectives can be accomplished (i.e. shorter or longer queues on the on ramps) Finally a numerical metering rate is estimated g iven all of the rule outcomes. Just as the inputs to the controller are represented by fuzzy classes to translate from a numerical input to a set of linguistic variables, so is the metering rate represented by a set of fuzzy classes to convert from a set of linguistic variab les to a single metering rate. This reverse process from a fuzzy to a crisp set is known as de fuzzification.

PAGE 46

46 3. 3 Scenarios The scope of this study encompasses the assessment of a number of simu lated traffic control scenarios. The entire set of scenarios is inherently div ided in to four different subsets. Initially traffic operations on the I 95 network are examined under the no traffic control scheme and for three different traffic demand levels. This subset of scena rios is presented in Table 3 7. The second subset include s those scenarios that only test the performance of the individual operation of the VSL algorithms (Table 3 8). Both VSL algorithms are evaluated for different number and locations of the VSL signs, and varying sensitivity of their operation (i.e. occupanc y and volume thresholds) at th e three demand levels. The third subset of scenarios considers only the simulation of the two ramp metering algorithms. The testing is also done for the same abovementioned demand levels and for different values of the ALINEA regulator parameter (i.e. K R ) and the fuzzy rule weights. The characteristics of these scenarios are shown in Table 3 9 The final subset is comprised by all the scenarios that correspond to a combination of a ramp metering (i.e. ALINEA or Fuzzy Logic) and a VSL algorithm (i.e Occupancy or Volume based). The four potential combinations are evaluated at three different traffic demand levels (i.e. Low, Medium and High demand case) and for different settings of the operational characteristics of the respective algorithms. The joint operation of the ALINEA algorithm either with the occupancy based VSL algorithm or the volume based one, is examined for different values of the ALINEA regulator parameter K R ; different number and locations of the VSL signs and varying sensitivity of the VSL algorithm s (i.e. occupancy and volume thresholds). The combined operation of the Fuzzy Logic algorithm either with the occupancy based VSL algorithm or the

PAGE 47

47 volume based one, is assessed for different values of the fuzzy rule weights, different number and locations of the VSL signs, and varying sensitivity of the VSL algorithms (i.e. o ccupancy and volume thresholds) These scenarios are presented in Tables 3 10 to 3 1 3 The medium traffic demand level corresponds to the d emand in the calibrated I 95 in Miami network. This is considered as the base case demand scenario and it yields heavily congested conditions on the network. The low demand case is obtained by decreasing the entering traffic form NW 125 th St. and from Opa Locka Blvd. by 20% for every simulated time period (i.e. 15 minute intervals). On the other hand, the high demand level is obtained by increasing incoming traffic from these two local arterials by 20% for each simulated time period (i.e. 15 minute inte rval s). The ALINEA ramp metering algorithm will be simulated for two different values of the regulator parameter K R The first value (i.e. K R = 1.17 veh/min ) has been selected according to results from field applications of the algorithm, while the second (i.e K R = 0.32 veh/min Logic ramp metering algorithm two different sets of values for the fuzzy rule weights have been implemented. The first set corresponds to the values proposed by Meld rum and Taylor (2000) and are presented in Table 3 6, while the modified values shown in Table 3 14 have been chosen to evaluate the performance of the algorithm when it becomes more sensitive to the on ramp traffic. For every scenario that examines the pe rformance of a VSL algorithm, the detector station that collects the traffic counts which determine its respective operation is installed 0.5 miles downstream of the entry point from the HOT lanes to the GP lanes

PAGE 48

48 on the northbound direction of I 95 in Miam i. This location has b een selected as it is the point where one of the occurring breakdown s i s initiating in the network. At each demand level the VSL algorithms are tested when: one VSL sign is installed 0.5 miles upstream of the bottleneck location one VSL sign is installed 1.0 mile upstream of the bottleneck location two VSL signs are installed; the first 1.0 mile upstream of the bottleneck location and the second 2.0 miles upstream of the bottleneck location. 3. 4 Analysis Process The objective of this analysis is twofold. Firstly to evaluate the effects of the combined operation of ramp metering and VSL on the performance of a freeway facility and secondly to identify any possible interactions of the two systems when they are concurrently implemented Th ese will be achieved by comparing network wide and link statistics between the integrated non integrated and no control scenarios. The network wide statistics will be presented in tabulated format, while t he link statistics will be presented in the for m of diagrams for every scenario The network wide statistics that will be provided are: Average network speed (mph) throughout the entire simulation Total distance traveled (miles) by all the vehicles during the simulation Total travel time (hours) The link statistics that will be presented for every scenario are: Average speed (mph) per link for every time period (i.e. 15 minute intervals) Throughput (vphpl) upstream and downstream of the bottleneck locations during the entire simulation time line. Ave rage delay (min) per vehicle behind two metered ramp s for every time period (i.e. 15 minute intervals) Average travel time (min) per vehicle to cross two predetermined route s within the network for every time period (15 minute intervals).

PAGE 49

49 Ramp metering ra tes ( veh/min ) at one metered ramp throughout the entire simulation. VSL rates (mph) for every VSL sign during the entire simulation. This analysis will evaluate the joint operation of ramp metering and VSL compared to the ramp metering only, VSL only and n o traffic control case. This evaluation will indicate whether the integrated control scheme is superior to more simple traffic management strategies. Moreover, it will identify which combination of ramp metering and VSL algorithm performs better compared t o the rest of the combinations, and for which settings of the respective algorithms (i.e. ALINEA regulator parameter, fuzzy rule weight, number of VSL signs, location of VSL signs etc.). As this analysis has been conducted for different traffic demand leve ls, it will also provide guidelines for the appropriate implementation of a traffic control strategy under different prevailing traffic conditions. The individual operation of each ramp metering and VSL a lgorithm will be also assessed. The assessment will reveal how the different aspects of the algorithms should be adjusted so that their performance is improved when they are applied alone. The analysis will also show the i nteractions between the VSL and the ramp metering system s. This interaction will be e valuated based on ramp metering rates under various VSL scenarios, and also based on the posted speed limits as a function of the ramp metering algorithm implemented. It is expected that the analysis will reveal the extent to which the one system can compl ement the other in order to regulate the traffic flow more efficiently.

PAGE 50

50 Table 3 1. Occupancy thresholds for displayed speed limits (Scenario 1). Traffic Category Occupancy Threshold for decreasing Speed Limit (%) Occupancy Threshold for increasing Speed Limit (%) Speed Limit (mph) Free Flow <16 <12 50 Light Congestion 16 28 12 25 45 Heavy Congestion >28 >25 40 Table 3 2. Occupancy thresholds for displayed speed limits (Scenario 2). Traffic Category Occupancy Threshold for decreasing Speed Limit (%) Occupancy Threshold for increasing Speed Limit (%) Speed Limit (mph) Free Flow <10 <8 50 Light Congestion 10 30 8 27 45 Heavy Congestion >30 >27 40 Table 3 3. Volume thresholds for displayed speed limits (Scenario 1). Flow Threshold for decreasing Speed Limit (vphpl) Flow Threshold for increasing Speed Limit (vphpl) Speed Limit (mph) < 1650 50 > 1650 < 1450 45 > 2050 < 1850 40 Table 3 4. Volume thresholds for displayed speed limits (Scenario 2). Flow Threshold for decreasing Speed Limit (vphpl) Flow Threshold for increasing Speed Limit (vphpl) Speed Limit (mph) < 1450 50 > 1450 < 1250 45 > 2000 < 1800 40

PAGE 51

51 T able 3 5. Description of fuzzy logic ramp metering a lgorithm i nputs Input Typical Detector Locations Local Occupancy Mainline Station just upstream of merge Local Speed Same as for Local Occupancy Downstream Occupancy Multiple downstream stations Downstream Speed Same as for downstream occupancy Queue Occupancy Queue detector on the ramp Advance Queue Occupancy Tail end of the available queue storage T able 3 6. Rule b ase for fuzzy ramp m etering a lgorithm Rule Rule Weight Rule Premise Rule Outcome 1 2.5 If local occupancy is VB Metering rate is VS 2 1.0 If local occupancy is B Metering rate is S 3 1.0 If local occupancy is M Metering rate is M 4 1.0 If local occupancy is S Metering rate is B 5 1.0 If local occupancy is VS Metering rate is VB 6 3.0 If local speed is VS AND local occupancy is VB Metering rate is VS 7 1.0 If local speed is S Metering rate is S 8 1.0 If local speed is B Metering rate is B 9 1.0 If local speed is VB AND local occupancy is VS Metering rate is VB 10 4.0 If downstream speed is VS AND downstream occupancy is VB Metering rate is VS 11 2.0 If queue occupancy is VB Metering rate is VB 12 4.0 If advance queue occupancy is VB Metering rate is VB

PAGE 52

52 Table 3 7. Scenarios evaluating traffic on the I 95 in Miami when no traffic management scheme is implemented. Scenario # Demand (veh/hr) Ramp Metering Algorithm VSL Algorithm 1 A Low B Medium C High

PAGE 53

53 Table 3 8 Scenarios to evaluate the operation of the proposed VSL algorithms. Scenario # Demand (veh/hr) VSL Algorithm Number of VSL signs Locations of VSL signs VSL operation 2 A Low Volume Based One 0.5 miles upstream Scenario 1 B Low Volume Based One 0.5 miles upstream Scenario 2 C Low Volume Based One 1.0 mile upstream Scenario 1 D Low Volume Based Two 1.0 mile upstream Scenario 1 E Low Occupancy Based One 0.5 miles upstream Scenario 1 F Low Occupancy Based One 0.5 miles upstream Scenario 2 G Low Occupancy Based One 1.0 mile upstream Scenario 1 H Low Occupancy Based Two 1.0 mile upstream Scenario 1 3 A Medium Volume Based One 0.5 miles upstream Scenario 1 B Medium Volume Based One 0.5 miles upstream Scenario 2 C Medium Volume Based One 1.0 mile upstream Scenario 1 D Medium Volume Based Two 1.0 mile upstream Scenario 1 E Medium Occupancy Based One 0.5 miles upstream Scenario 1 F Medium Occupancy Based One 0.5 miles upstream Scenario 2 G Medium Occupancy Based One 1.0 mile upstream Scenario 1 H Medium Occupancy Based Two 1.0 mile upstream Scenario 1 4 A High Volume Based One 0.5 miles upstream Scenario 1 B High Volume Based One 0.5 miles upstream Scenario 2 C High Volume Based One 1.0 mile upstream Scenario 1 D High Volume Based Two 1.0 mile upstream Scenario 1 E High Occupancy Based One 0.5 miles upstream Scenario 1 F High Occupancy Based One 0.5 miles upstream Scenario 2 G High Occupancy Based One 1.0 mile upstream Scenario 1 H High Occupancy Based Two 1.0 mile upstream Scenario 1

PAGE 54

54 Table 3 9 Scenarios evaluating the operation of the ALINEA and the Fuzzy Logic Ramp Metering Algorithms. Scenario # Demand (veh/hr) Ramp Metering Algorithm ALINEA Calibration Parameter (veh/min) Fuzzy Rule Weights 5 A Low ALINEA 0.32 B Low ALINEA 1.17 C Low Fuzzy Logic Proposed D Low Fuzzy Logic Modified 6 A Medium ALINEA 0.32 B Medium ALINEA 1.17 C Medium Fuzzy Logic Proposed D Medium Fuzzy Logic Modified 7 A High ALINEA 0.32 B High ALINEA 1.17 C High Fuzzy Logic Proposed D High Fuzzy Logic Modified

PAGE 55

55 Table 3 10. Scenarios evaluating the joint operation of the ALINEA ramp metering and the volume based VSL algorithms. Scenario # Demand (veh/hr) ALINEA Calibration Parameter (veh/min) Number of VSL signs Locations of VSL signs VSL operation 8 A Low 0.32 One 0.5 miles upstream Scenario 1 B Low 1.17 One 0.5 miles upstream Scenario 1 C Low 0.32 One 0.5 miles upstream Scenario 2 D Low 0.32 One 1.0 mile upstream Scenario 1 E Low 0.32 Two 1.0 mile upstream Scenario 1 9 A Medium 0.32 One 0.5 miles upstream Scenario 1 B Medium 1.17 One 0.5 miles upstream Scenario 1 C Medium 0.32 One 0.5 miles upstream Scenario 2 D Medium 0.32 One 1.0 mile upstream Scenario 1 E Medium 0.32 Two 1.0 mile upstream Scenario 1 10 A High 0.32 One 0.5 miles upstream Scenario 1 B High 1.17 One 0.5 miles upstream Scenario 1 C High 0.32 One 0.5 miles upstream Scenario 2 D High 0.32 One 1.0 mile upstream Scenario 1 E High 0.32 Two 1.0 mile upstream Scenario 1

PAGE 56

56 Table 3 11 Scenarios evaluating the joint operation of the ALINEA ramp metering and the occupancy based VSL algorithms. Scenario # Demand (veh/hr) ALINEA Calibration Parameter (veh/min) Number of VSL signs Locations of VSL signs VSL operation 1 1 A Low 0.32 One 0.5 miles upstream Scenario 1 B Low 1.17 One 0.5 miles upstream Scenario 1 C Low 0.32 One 0.5 miles upstream Scenario 2 D Low 0.32 One 1.0 mile upstream Scenario 1 E Low 0.32 Two 1.0 mile upstream Scenario 1 1 2 A Medium 0.32 One 0.5 miles upstream Scenario 1 B Medium 1.17 One 0.5 miles upstream Scenario 1 C Medium 0.32 One 0.5 miles upstream Scenario 2 D Medium 0.32 One 1.0 mile upstream Scenario 1 E Medium 0.32 Two 1.0 mile upstream Scenario 1 1 3 A High 0.32 One 0.5 miles upstream Scenario 1 B High 1.17 One 0.5 miles upstream Scenario 1 C High 0.32 One 0.5 miles upstream Scenario 2 D High 0.32 One 1.0 mile upstream Scenario 1 E High 0.32 Two 1.0 mile upstream Scenario 1

PAGE 57

57 Table 3 12. Scenarios evaluating the joint operation of the Fuzzy Logic ramp metering and the volume based VSL algorithms. Scenario # Demand (veh/hr) Fuzzy Rule Weights Number of VSL signs Locations of VSL signs VSL operation 1 4 A Low Proposed One 0.5 miles upstream Scenario 1 B Low Modified One 0.5 miles upstream Scenario 1 C Low Proposed One 0.5 miles upstream Scenario 2 D Low Proposed One 1.0 mile upstream Scenario 1 E Low Proposed Two 1.0 mile upstream Scenario 1 1 5 A Medium Proposed One 0.5 miles upstream Scenario 1 B Medium Modified One 0.5 miles upstream Scenario 1 C Medium Proposed One 0.5 miles upstream Scenario 2 D Medium Proposed One 1.0 mile upstream Scenario 1 E Medium Proposed Two 1.0 mile upstream Scenario 1 1 6 A High Proposed One 0.5 miles upstream Scenario 1 B High Modified One 0.5 miles upstream Scenario 1 C High Proposed One 0.5 miles upstream Scenario 2 D High Proposed One 1.0 mile upstream Scenario 1 E High Proposed Two 1.0 mile upstream Scenario 1

PAGE 58

58 Table 3 13 Scenarios evaluating the joint operation of the Fuzzy Logic ramp metering and the occupancy based VSL algorithms. Scenario # Demand (veh/hr) Fuzzy Rule Weights Number of VSL signs Locations of VSL signs VSL operation 1 7 A Low Proposed One 0.5 miles upstream Scenario 1 B Low Modified One 0.5 miles upstream Scenario 1 C Low Proposed One 0.5 miles upstream Scenario 2 D Low Proposed One 1.0 mile upstream Scenario 1 E Low Proposed Two 1.0 mile upstream Scenario 1 1 8 A Medium Proposed One 0.5 miles upstream Scenario 1 B Medium Modified One 0.5 miles upstream Scenario 1 C Medium Proposed One 0.5 miles upstream Scenario 2 D Medium Proposed One 1.0 mile upstream Scenario 1 E Medium Proposed Two 1.0 mile upstream Scenario 1 1 9 A High Proposed One 0.5 miles upstream Scenario 1 B High Modified One 0.5 miles upstream Scenario 1 C High Proposed One 0.5 miles upstream Scenario 2 D High Proposed One 1.0 mile upstream Scenario 1 E High Proposed Two 1.0 mile upstream Scenario 1

PAGE 59

59 Table 3 14. Modified rule weights of fuzzy ramp metering algorithm rule base. Rule Rule Weight Rule Premise Rule Outcome 1 1.0 If local occupancy is VB Metering rate is VS 2 1.0 If local occupancy is B Metering rate is S 3 1.0 If local occupancy is M Metering rate is M 4 1.0 If local occupancy is S Metering rate is B 5 1.0 If local occupancy is VS Metering rate is VB 6 1 .0 If local speed is VS AND local occupancy is VB Metering rate is VS 7 1.0 If local speed is S Metering rate is S 8 1.0 If local speed is B Metering rate is B 9 1.0 If local speed is VB AND local occupancy is VS Metering rate is VB 10 4.0 If downstream speed is VS AND downstream occupancy is VB Metering rate is VS 11 6 .0 If queue occupancy is VB Metering rate is VB 12 6 .0 If advance queue occupancy is VB Metering rate is VB

PAGE 60

60 CHAPTER 4 THE SIMULATION NETWO RK FOR I 95 IN MIAMI 4.1 The Study Site I 95 in Miami was selected as the simulation network for the evaluation of the joint operation of ramp metering and VSL and their respective interactions. This network had already been simulated in CORSIM by the F lorida D epartment of T ransportation (FDOT) during the process of the HOT lanes design in Miami and was provided to the Transportation Research Center within the scope of the Managed Lane Operations project However, the objectives of this work required the reconfiguration and calibration of the ini tial CORSIM files provided by FDOT to facilitate the current analysis The following sections describe the configuration of the network and the calibration process to ensure that the simulator accurately reflects field conditions before experiments are con ducted. 4.2 Configuration of the Network Several changes were made to the initial CORSIM files to ensure that traffic conditions on the network could be accurately replicated and that the objectives of this study could be accomplished The changes implemented are related to the geometry and the extent of the study area, the volumes entering the network, the implementation of the ramp metering algorithm, and the simulated detector locations used in calibrating the model. The remainder of this section describes these four types of changes. 4.2.1 Network Geometry The initial CORSIM file provided by FDOT was replicating the traffic conditions on both directions of I 95 in Miami. The file included the interchanges of the Expressway with I 395, I 195 and all the local arterials connecting the urban streets with the freeway

PAGE 61

61 network through ramps. Part of the Turnpike was also included in the network. Thus this original file was very extensive and required extensive computational resources (each run needed a pproximatel y 45 minutes to be completed). As the objective s of this research did not require the simulation of all these roadway sections the network was modified to replicate the operation of northbound I 95 in Miami together with all the ramps merging a n d diverging from the freeway. These segments were sufficient for evaluating the integrated operation of ramp metering and VSL and their respective interactions Operations on I 395, I 195, the Turnpike and local arterials were not important for the purpose s of this project, and only their connections with I 95 were ke pt and modified appropriately. resour ces on the freeway operations. As an example, Figure 4 1 illustrates the geometry of the NW 62 n d Street interchange before the changes, and Figure 4 2 illustrates the same interchange after the changes were implemented. 4.2.2 Incoming Volumes Field data from the STEWARD database were used to obtain incoming volumes etwork. Volu me data from the 7 th of October 2009 were extracted from the database and were input into CORSIM. T he quality and amount of data from that day are sufficient compared to those of nearby dates. Data from the time period between 15:30 18:30 pm of the select ed day were used. That time represents the afternoon peak period on the northbound direction. Volumes were obtained in 15 minute intervals, and the simulation contains a total of 12 analysis periods. The volumes input at the entry nodes were mainly derived from data provided from the inductive loop detectors at the on ramps. For on ramps where loop detector data were not available, data from two detectors (one upstream of the merge point and

PAGE 62

62 one downstream) were utilized and the entry volume was estimated a s the difference between the two counts. At diverge points, the amount of traffic exiting from the mainline was calculated either by using the upstream and downstream detectors as mentioned above, or by keeping the same percentage of exiting traffic as spe cified in the CORSIM files provided by FDOT. The simulation network also includes the HOT lanes of the northbound direction. The split of traffic between HOT and general purpose lanes was derived from the first station located just upstream of the diverge point north of NW 62 nd St. Appendix A provides the relevant data regarding the loading of the network (mainline traffic volumes, entering traffic and exiting traffic at entry and exit nodes respectively ) 4.2.3 Implementation of the Ramp Metering Algorith m In the original CORSIM file, traffic entering from the on ramps onto the I 95 in Miami was not reg ulated by any control strategy. Thus, it was necessary to replicate the ramp metering algorithm in effect at the time the data were collected (October, 2009). This algorithm, which is a pre timed, fixed rate algorithm, can e asily be replicated in CORSIM. Therefore, the metering rates (Table 4 1) as provided by the Sunguide Transportation Management Center in Miami were coded within CORSIM 4.2.4 Detector Placement t hroughout the Network Calibration involves the comparison of traffic operations in the field to the simulated network and adjustment of the simulation as necessary. To complete the process of calibration it is necessary to install detectors in the field, and collect various performance measures related to traffic operations. The detectors that are supplying the STEWARD database with data are providing measurements of volume, speed and

PAGE 63

63 occupancy at particul ar locations along the network. Thus, fo r calibration and validation purposes eleven detectors were installed throughout the CORSIM network at locations that correspond to those of the actual detectors in the field. The first eight detectors were set up along the General Purpose Lanes, while the l ast three ones along the HOT Lanes. The actual and simulated locations of those detectors are shown in Figure 4 3 and Table 4 2 4.3 Calibration P rocess Calibration is the process of model adjustment to ensure that the network performs accurately and tha t the assumptions made in develo ping the model are reasonable. Calibration involves a comparison of selected performance measures between the simulate d corridor and the field data. For the purposes of this study network volumes and speeds were monitored t hroughout the simulated network to identify potential significant differences at specific locations. Since CORSIM is a stochastic simulator, it uses random number generators to replicate traffic conditions, and each run should be viewed as one sample of t he experiment. Several runs of the simulator are needed to obtain an estimate of the Thus, after the completion of the modifications outlined in the previous section, the required number of runs was determined. Initiall y, the model was executed 10 times, and the average network speed was obtained (Table 4 3 ). Based on these results after assuming an allowable error of e =0.05 mph and a 95% confidence level, the required number of runs was estimated to be 7. Thus, in sub sequent analysis 10 runs will be conducted, whic h are more than adequate for this context

PAGE 64

64 The reconciliation of field and simulated traffic counts was achieved after an extensive experimentation process with the available calibration parameters in CORSIM. The adjustments to the values of these parameters were finalized through trial and error after several iterations. The calibration parameter with the most profound effect on the simulated traffic operations was found to be the car following sensit ivity fa ctor. This factor significantly influences the desired time headway during car following. Therefore, this factor was the one to be mostly modified for the calibration of the network. Appendix A includes the value of that factor per network link, together with the free flow speed at each link. Parameters affecting the lane changing activity, the start up delay of vehicles in front of meters, and the arrival rate of traffic into the net work were also adjusted. Tables 4 4 to 4 15 and Figures 4 4 to 4 27 compare the field counts to those obtained in the simulation after the completion of the calibration. Simulated speeds deviate from field measurements more than volumes, but they are still very close to fiel d conditions at most instances. Location 4 breaks down earlier in the simulation than in the field, causing the queue to dissipate earlier as well. This discrepancy should be reality. For example, in CORSIM drivers do not adjust their speed to reduce speed differential between lanes. Drivers in the field are more prone to consider their broader environment and be affected by it. Generally, in the simulation the beginning and b uild up of congestion follows a similar pattern to that observed in the field. The calibration was based on both traffic speeds and vol umes over multiple time periods Simulated mean speeds deviate about

PAGE 65

65 7.00 mph from the field observations, while traffic volumes around 100 vehicles per 15 minutes from the corresponding field data Therefore, it was concluded that the simulated network is replicating actual traffic conditions sufficiently

PAGE 66

66 Table 4 1. Location and metering rate of each ramp signal along I 95 Location Metering Rate (veh/min) 1 I 95 N orth of NW 62 S t. 20 2 I 95 at NW 69 S t. 20 3 I 95 N orth of NW 81 S t. 20 4 I 95 at NW 96 S t. 20 5 I 95 N orth of NW 103 S t. 20 6 I 95 S outh of NW 131 S t. 20 7 I 95 N orth of OPA LOCKA B lvd. 24 8 I 95 S outh of US 441 24 Table 4 2. Location of field and simulated detectors Location Actual Station ID# CORSIM Station ID# 1 I 95 North of NW 17 S t. 600291 1 2 I 95 N orth of NW 62 S t. 600471 2 3 I 95 N orth of NW 77 S t. 600521 3 4 I 95 at NW 96 S t. 600621 4 5 I 95 S outh of NW 11 1 S t. 600711 5 6 I 95 S outh of NW 131 S t. 600791 6 7 I 95 S outh of NW 15 1 S t. 600921 7 8 I 95 S outh of US 441 600981 8 9 I 95 N orth of NW 62 S t. 690471 9 10 I 95 at NW 96 S t. 690621 10 11 I 95 S outh of NW 131 S t. 690791 11

PAGE 67

67 Table 4 3. Average network speed for each simulation run Run # Average Network Speed (mph) 1 49.13 2 49.20 3 49.09 4 49.12 5 49.25 6 49.26 7 49.13 8 49.11 9 49.13 10 49.04 Table 4 4. Field and simulated speeds and volumes 1 st time period 1 st Time period (15:30 15:45) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 42.27 53.16 796 783 I 95 North of NW 62 S t. 30.55 22.67 1607 1622 I 95 North of NW 77 S t. 31.35 39.30 1539 1660 I 95 at NW 96 S t. 40.32 37.32 1570 1588 I 95 South of NW 11 1 S t. 35.07 41.76 1675 1711 I 95 S outh of NW 131 S t. 51.96 46.66 1529 1607 I 95 South of NW 15 1 S t. 54.70 43.72 2218 2114 I 95 S outh of US 441 54.70 57.92 861 839 I 95 N orth of NW 62 S t. 63.42 66.88 514 609 I 95 at NW 96 S t. 66.94 65.98 491 611 I 95 S outh of NW 131 S t. 62.55 65.32 586 606

PAGE 68

68 Table 4 5. Field and simulated speeds and volumes 2 nd time period 2 nd Time period (15: 45 1 6 : 00 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 48.04 52.82 851 862 I 95 North of NW 62 S t. 32.10 31.50 1642 1749 I 95 North of NW 77 S t. 33.81 40.69 1465 1630 I 95 at NW 96 S t. 42.54 24.47 1602 1552 I 95 South of NW 11 1 S t. 37.83 41.73 1741 1712 I 95 S outh of NW 131 S t. 46.37 46.78 1474 1576 I 95 South of NW 15 1 S t. 54.98 43.74 2107 2042 I 95 S outh of US 441 55.41 58.09 832 804 I 95 N orth of NW 62 S t. 61.28 67.27 531 544 I 95 at NW 96 S t. 65.36 66.31 440 554 I 95 S outh of NW 131 S t. 59.53 65.90 549 564 Table 4 6. Field and simulated speeds and volumes 3 rd time period 3 rd Time period (1 6 : 00 1 6 : 15 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 36.12 52.40 862 932 I 95 North of NW 62 S t. 43.75 50.02 1634 1656 I 95 North of NW 77 S t. 35.15 45.89 1521 1567 I 95 at NW 96 S t. 44.13 23.36 1651 1545 I 95 South of NW 11 1 S t. 38.38 41.76 1790 1721 I 95 S outh of NW 131 S t. 49.94 46.93 1588 1565 I 95 South of NW 15 1 S t. 44.75 40.76 2292 1993 I 95 S outh of US 441 55.64 58.30 890 740 I 95 N orth of NW 62 S t. 63.05 66.41 616 681 I 95 at NW 96 S t. 68.47 65.58 516 661 I 95 S outh of NW 131 S t. 62.52 65.21 610 638

PAGE 69

69 Table 4 7. Field and simulated speeds and volumes 4 th time period 4 th Time period (1 6 : 15 1 6 : 30 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 57.84 53.16 813 799 I 95 North of NW 62 S t. 53.83 50.95 1616 1570 I 95 North of NW 77 S t. 36.35 45.59 1568 1585 I 95 at NW 96 S t. 42.70 21.87 1601 1569 I 95 South of NW 11 1 S t. 36.45 41.58 1742 1709 I 95 S outh of NW 131 S t. 49.92 46.96 1446 1553 I 95 South of NW 15 1 S t. 31.01 27.03 1912 1972 I 95 S outh of US 441 55.93 58.44 766 725 I 95 N orth of NW 62 S t. 63.07 66.75 619 636 I 95 at NW 96 S t. 68.45 65.91 481 637 I 95 S outh of NW 131 S t. 63.07 65.51 573 650 Table 4 8. Field and simulated speeds and volumes 5 th time period 5 th Time period (1 6 : 30 1 6 : 45 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 58.83 53.11 807 799 I 95 North of NW 62 S t. 56.61 51.57 1560 1509 I 95 North of NW 77 S t. 56.80 52.58 1505 1456 I 95 at NW 96 S t. 41.86 24.72 1616 1586 I 95 South of NW 11 1 S t. 36.60 41.56 1725 1720 I 95 S outh of NW 131 S t. 36.33 46.82 1346 1582 I 95 South of NW 15 1 S t. 28.75 30.70 1928 1935 I 95 S outh of US 441 55.31 58.43 776 733 I 95 N orth of NW 62 S t. 62.57 67.24 620 585 I 95 at NW 96 S t. 68.08 66.27 518 602 I 95 S outh of NW 131 S t. 62.64 65.81 549 608

PAGE 70

70 Table 4 9. Field and simulated speeds and volumes 6 th time period 6 th Time period (1 6 : 45 1 7 : 00 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 57.31 52.68 949 917 I 95 North of NW 62 S t. 56.86 51.32 1622 1554 I 95 North of NW 77 S t. 56.20 53.88 1555 1486 I 95 at NW 96 S t. 43.97 46.42 1583 1495 I 95 South of NW 11 1 S t. 37.90 41.85 1747 1722 I 95 S outh of NW 131 S t. 28.18 40.08 1340 1731 I 95 South of NW 15 1 S t. 24.57 28.74 1799 2059 I 95 S outh of US 441 55.31 58.02 790 875 I 95 N orth of NW 62 S t. 63.51 66.95 636 623 I 95 at NW 96 S t. 67.58 65.91 504 616 I 95 S outh of NW 131 S t. 61.16 65.67 571 605 Table 4 10. Field and simulated speeds and volumes 7 th time period 7 th Time period (1 7 : 00 1 7 : 15 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 46.69 52.05 926 986 I 95 North of NW 62 S t. 56.39 50.99 1649 1604 I 95 North of NW 77 S t. 55.07 47.97 1603 1580 I 95 at NW 96 S t. 50.01 52.49 1563 1552 I 95 South of NW 11 1 S t. 33.36 42.02 1565 1709 I 95 S outh of NW 131 S t. 23.43 33.98 1221 1739 I 95 South of NW 15 1 S t. 26.62 33.26 1680 2097 I 95 S outh of US 441 56.25 57.77 681 939 I 95 N orth of NW 62 S t. 64.76 66.93 637 589 I 95 at NW 96 S t. 69.99 66.34 533 588 I 95 S outh of NW 131 S t. 64.79 65.71 564 594

PAGE 71

71 Table 4 11. Field and simulated speeds and volumes 8 th time period 8 th Time period (1 7 : 15 1 7 : 30 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 44.58 52.45 927 923 I 95 North of NW 62 S t. 56.40 50.49 1719 1679 I 95 North of NW 77 S t. 49.17 40.49 1620 1614 I 95 at NW 96 S t. 31.85 52.57 1512 1638 I 95 South of NW 11 1 S t. 28.68 42.28 1476 1700 I 95 S outh of NW 131 S t. 21.68 30.12 1201 1688 I 95 South of NW 15 1 S t. 25.04 29.61 1679 2041 I 95 S outh of US 441 56.63 58.08 698 825 I 95 N orth of NW 62 S t. 64.81 67.21 582 527 I 95 at NW 96 S t. 70.98 66.37 436 536 I 95 S outh of NW 131 S t. 67.20 65.96 545 550 Table 4 12. Field and simulated speeds and volumes 9 th time period 9 th Time period (1 7 : 30 1 7 : 45 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 54.27 52.38 924 952 I 95 North of NW 62 S t. 53.97 49.91 1598 1596 I 95 North of NW 77 S t. 35.80 44.35 1407 1593 I 95 at NW 96 S t. 30.04 46.00 1396 1641 I 95 South of NW 11 1 S t. 31.10 41.52 1485 1702 I 95 S outh of NW 131 S t. 22.94 20.43 1114 1603 I 95 South of NW 15 1 S t. 28.83 27.77 1843 2023 I 95 S outh of US 441 55.74 57.94 750 859 I 95 N orth of NW 62 S t. 65.41 67.08 625 596 I 95 at NW 96 S t. 67.57 66.32 474 580 I 95 S outh of NW 131 S t. 64.88 65.69 530 567

PAGE 72

72 Table 4 13. Field and simulated speeds and volumes 10 th time period 10 th Time period (1 7 : 45 1 8 : 00 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 64.87 52.46 885 926 I 95 North of NW 62 S t. 56.20 51.10 1540 1511 I 95 North of NW 77 S t. 38.10 53.80 1339 1401 I 95 at NW 96 S t. 31.23 45.15 1371 1479 I 95 South of NW 11 1 S t. 27.97 40.32 1460 1700 I 95 S outh of NW 131 S t. 31.79 16.60 1458 1463 I 95 South of NW 15 1 S t. 34.02 27.85 2055 2013 I 95 S outh of US 441 56.10 57.80 811 939 I 95 N orth of NW 62 S t. 65.74 67.63 559 492 I 95 at NW 96 S t. 71.00 66.70 397 514 I 95 S outh of NW 131 S t. 63.72 65.88 503 531 Table 4 14. Field and simulated speeds and volumes 11 th time period 11 th Time period (1 8 : 00 1 8 : 15 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 65.16 52.29 909 927 I 95 North of NW 62 S t. 55.01 51.28 1550 1506 I 95 North of NW 77 S t. 34.39 53.56 1425 1454 I 95 at NW 96 S t. 42.19 51.34 1599 1530 I 95 South of NW 11 1 S t. 35.75 42.21 1694 1673 I 95 S outh of NW 131 S t. 42.22 27.42 1501 1474 I 95 South of NW 15 1 S t. 38.01 29.54 2143 2058 I 95 S outh of US 441 55.90 57.82 880 868 I 95 N orth of NW 62 S t. 66.20 67.08 563 569 I 95 at NW 96 S t. 70.55 66.60 427 560 I 95 S outh of NW 131 S t. 64.10 66.20 533 544

PAGE 73

73 Table 4 15. Field and simulated speeds and volumes 12 th time period 12 th Time period (1 8 : 15 1 8 : 30 ) Location Field Speeds Simulated Speeds Field Volumes Simulated Volumes I 95 North of NW 17 S t. 66.26 52.54 873 908 I 95 North of NW 62 S t. 57.06 51.29 1467 1487 I 95 North of NW 77 S t. 62.20 54.29 1363 1421 I 95 at NW 96 S t. 55.30 52.34 1526 1544 I 95 South of NW 11 1 S t. 41.54 41.43 1698 1736 I 95 S outh of NW 131 S t. 51.56 46.52 1426 1259 I 95 South of NW 15 1 S t. 48.08 43.83 2074 1816 I 95 S outh of US 441 56.42 58.27 889 774 I 95 N orth of NW 62 S t. 64.09 67.10 483 539 I 95 at NW 96 S t. 71.51 66.75 400 537 I 95 S outh of NW 131 S t. 64.41 66.00 503 547

PAGE 74

74 F igure 4 1. The I 95 Expressway at NW 62 nd St reet before the modifications F igure 4 2. The I 95 Expressway at NW 62 nd St reet after the modifications

PAGE 75

75 F igure 4 3. Location of detectors used for calibration on I 95 in Miami.

PAGE 76

76 Figure 4 4. Field and simulated speeds 1 st time period F igure 4 5. Field and simulated volumes 1 st time period 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 77

77 Figure 4 6. Field and simulated speeds 2 nd time period F igure 4 7. Field and s imulated v olumes 2 nd time p eriod 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 78

78 Figure 4 8. Field and simulated speeds 3 rd time period Figure 4 9. Field and simulated volumes 3 rd time period 0.00 10.00 20.00 30.00 40.00 50.00 60.00 70.00 80.00 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 79

79 Figure 4 10. Field and simulated speeds 4 th time period Figure 4 11. Field and simulated volumes 4 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 80

80 Figure 4 12. Field and simulated speeds 5 th time period Figure 4 13. Field and simulated volumes 5 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 81

81 Figure 4 14. Field and simulated speeds 6 th time period Figure 4 15. Field and simulated volumes 6 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 82

82 Figure 4 16. Field and simulated speeds 7 th time period F igure 4 17. Field and simulated volumes 7 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 83

83 F igure 4 18. Field and simulated speeds 8 th time period F igure 4 19. Field and simulated volumes 8 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 84

84 F igure 4 20. Field and simulated speeds 9 th time period F igure 4 21. Field and simulated volumes 9 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 85

85 Figure 4 22. Field and simulated speeds 10 th time period Figure 4 23. Field and simulated volumes 10 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 86

86 F igure 4 24. Field and simulated speeds 11 th time period F igure 4 25. Field and simulated volumes 11 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 87

87 Figure 4 26. Field and simulated speeds 12 th time period Figure 4 27. Field and simulated volumes 12 th time period 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 Speed (mph) CORSIM Station ID# Field Speeds Sim. Speeds 0 500 1000 1500 2000 2500 1 2 3 4 5 6 7 8 9 10 11 Volume (vph) CORSIM Station ID# Field Vol. Sim. Vol.

PAGE 88

88 CHAPTER 5 ANALYSIS OF THE SIMU LATION RESULTS 5.1 Traffic Impacts of the Simulated Traffic Management Schemes The analysis of t he simulation results is presented in this chapter. Network wide statistics (i.e. total distance traveled, total travel time, average network speed) and selected link statistics are shown for each examined scenario. The complete list of all the simulated s cenarios is provided in Tables 3 7 to 3 13. These results are studied to assess the effects that each traffic management scheme (i.e. integrated control, VSL only, ramp metering only, no control) exerts on traffic operations on the I 95 in Miami. The inter actions between ramp metering and VSL are also studied by examining the VSL rates together with the metering rates imposed at the on ramps which are located in the proximity of the VSL signs. The network wide and link statistics are discussed for every set of scenarios that corresponds to a different traffic control scheme and for all of the tested demand levels. Initially traffic conditions on the I 95 are described for the no control case. Then the impacts of VSL on traffic flow will be presented, followe d by those of the ramp metering only scenario. Afterwards, the results relevant to the integrated traffic management scheme (i.e. ramp metering and VSL) will be analyzed. A comprehensive list of the link statistics to be examined has been already provided in Chapter 3. More specifically, the space mean speed per link will be depicted for five different time periods (i.e. 4 th 6 th 8 th 10 th and 12 th ) for a number of the total scenarios. These time periods have been selected because they clearly portray the progression of congestion. Throughput was evaluated North of Opa Locka Boulevard and North of Biscayne Canal, because congestion typically begins around these areas

PAGE 89

89 between 16:15 16:30 pm for the no control case. These two locations are upstream of the bottleneck source. Thus, it was also deemed necessary to evaluate throughput downstream of the bottleneck location. Throughput is obtained South of US 441 for the best implementation of each traffic control strategy during the medium demand scenario. Avera ge travel time per vehicle has been also computed for two origin destinations (O D). The first O D originates at the entry from I 195 EB/WB and ends at the off ramp to 95 th Street. This route is 2.8 miles long and a vehicle can complete it in 3.36 min assu ming it is traveling at 50 mph. The second O D begins at the entry from NW 81 st St. and stretches till the exit to North Golden Glades Drive. It is 4.8 miles long and can be traveled in 5.76 min at 50 mph. These two paths constitute the most congested port ion of the freeway network. Thus, the comparison of the average travel time on these paths under the different types of traffic control will also reveal the extent to which they improve or deteriorate traffic flow on the network. Finally, the delay per ve hicle has been obtained at two on ramps. The one is located immediately north of 125 th St. and the other north of Opa Locka Boulevard. They are both along the most congested portion of the network. The estimation of delay per vehicle at these two on ramps will help assess the performance of each algorithm relative to queue and delay at thes e ramps. 5.1 .1 No control Case As it has been already mentioned (i.e. Chapter 3) three demand scenarios were examined: low, medium, and high. During the low demand scenario the average network speed is 39.16 mph (Table 5 1). During the medium demand scenario, when the incoming traffic from 125 th St. and Opa Locka Boulevard increases by 20% for every

PAGE 90

90 time period, the average network speed decreases from 39.16 mph to 34.86 mph (Table 5 1). Traffic conditions on the I 95 in Miami significantly deteriorat e under th e maximum demand scenario with an average network speed of 31.98 mph (Table 5 1). This is a decrease of 2.88 mph compared to the medium demand level and 7.18 mph compared to the low demand level. At the low demand level two distinct breakdowns occur in the network between 16:15 16:30 pm. The first one begins south of NW 103 rd St., while the second one north of Biscayne Canal. The severity of the first breakdown is larger, sinc e the average speed drops to 10 mph at this location; at the seco nd location speed drops to 25 mph (Figure 5 1). These two breakdowns evolve separately till the end of the simulation. At this point congestion at the downstream bottleneck has dissipated (Figures 5 2 to 5 5). Throughput north of Biscayne Canal drops from 17 40 veh/hr/lane in the beginning of the simulation to 1490 veh/hr/lane at the onset of congestion (i.e. 4 th time period) (Figure 5 7). Travel time per vehicle reaches up to 9 min from I 195 EB/WB to 95 th St. during the most congested time period (i.e. 17:45 to 18:00 pm), and up to 14.5 min from 81 st St. to N. Golden Glades Drive (Figures 5 8 and 5 9). The travel time on both routes is much higher than the corresponding free flow speed travel time (i.e. 3.36 min and 5.76 min respectively). This observation is in agreement with the speed over location results discussed earlier and presented in Figures 5 1 to 5 5. Finally, no vehicles are delayed at the on ramps, since no ramp metering has been implemented at this case and congestion is only limited to the mainl ine network (Figures 5 10 and 5 11). At the medium demand level the two previously separate congested areas merge between 17:45 18:00 pm. The congested area now extends from north of Biscayne

PAGE 91

91 Canal till south of NW 62 nd St., and the average speed per lin k across this freeway stretch oscillates around 10 mph (Figure 5 15). Moreover, at the end of the simulation congestion is not dissolved north of Biscayne Canal (Figure 5 16). Throughput drops below 1500 veh/hr/lane north of Opa Locka Boulevard and below 1400 veh/hr/lane north of Biscayne Canal when congestion starts in this area (Figures 5 17 and 5 18). Travel time per vehicle increases on both examined routes. Between 17:45 18:00 pm average travel time hikes up to 12 min from I 195 EB/WB till 95 th St. and up to 20 min from 81 st St. to N. Golden Glades Drive. This is an increase of 3.0 0 min and 5.50 min respectively compared to the low demand scenario (Figures 5 20 and 5 2 1 ). However, despite the demand growth, traffic operations at the on ramps north o f 125 th St. and north of Opa Locka Blvd. are not affected. No vehicles are backed up on these ramps (Figures 5 2 2 and 5 2 3 ). The merging of the two distinct breakdowns (i.e. north of NW 103 rd St. and north of Biscayne Canal) occurs earlier in the simulatio n time line when the highest demand scenario is tested (i.e. between 17:15 17:30 pm instead of 17:45 18:00 pm) (Figure 5 2 6 ). Thus, the freeway section affected is longer during the 10 th time period (i.e. 17:45 18:00 pm) and extends from north of Biscayne Canal till south of NW 53 rd St. (Figure 5 2 7 ). Hence, by the end of the simulation the biggest portion of the freeway still remains heavily congested (Figure 5 2 8 ). Throughput north of Opa Locka Blvd. and north of Biscayne Canal is lower for the no control case compared to other control strategies between the 4 th and the 10 th time period at the high demand level (Figures 5 2 9 and 5 30 ). The average travel time required per vehicle to cr oss the freeway segment between I 195 EB/WB and 95 th St.

PAGE 92

92 has further increased up to 14.00 min during the 10 th time period (Figure 5 3 1 ). For the same time period travel time has reached 24.00 min per vehicle from 81 st St. to N. Golden Glades Dr. (Figure 5 3 2 ). Therefore, there is a difference of 5.00 min per vehicle for the first route and 9.50 min per vehicle for the second one compared to the low demand case. It is also critical to be mentioned that traffic flow is impeded at the on ramp north of 125 th S t., since demand has increased substantially. Each vehicle entering the freeway from this ramp is approximately delayed 0.6 min betwe en 16:45 17:15 pm (Figure 5 33 ). No queue is formed though at the on ramp north of Opa Locka Blvd. (Figure 5 3 4 ). The pre ceding detailed description of the traffic operations on Miami I 95 when no traffic control management strategy is implemented results to a number of general observations. As demand increases on Miami I 95 congestion occurs earlier on the network and the c ongested area becomes longer with time if traffic is un regulated. Moreover, queues do not dissipate by the end of the simulation time line at this case. These observations are also consistent with the depicted link performance measures. In the beginning of congestion throughput drops significantly at the bottleneck locations. Travel time along the most congested portions of the freeway is climbing up proportionally to the increasing demand. Even traffic conditions on ramps slightly worsen during the maximum demand scenario. Overall, traffic conditions on the Miami I 95 freeway network unavoidably deteriorate if demand substantially increases and no traffic control scheme is deployed. 5.1 .2 VSL only Case In this study two different types of VSL algorithms hav e been considered as described in Chapter 3; one adjusting the speed limits based on occupancy

PAGE 93

93 measurements, and one according to volume measurements. For both algorithms the VSL signs were installed upstream of Biscayne Canal, since congestion begins at t his location between 16:15 16:30 pm during the no control case. Installation of one sign was tested either 0.5 miles upstream of the bottleneck source or 1.0 mile upstream. Installation of two signs was also examined. At this case the first sign is locat ed 1.0 mile upstream from the bottleneck and the second 1.0 mile upstream from the first one. The occupancy and volume thresholds applied during the simulation of VSL are presented in Tables 3 1 to 3 4. During the low demand scheme the average network spee d is 40.10 mph when the occupancy based VSL algorithm is implemented (Table 5 2).This occurs when one VSL sign is installed upstream of the bottleneck, and the applied occupancy thresholds are those of Scenario 1 (Table 3 1). This is the highest average n etwork speed that can be attained by a specific implementation of the considered VSL algorithms at this demand level. It is 0.94 mph higher compared to the no control case. This small difference is also consistent with the information in the charts that p resent the average speed per link (Figures 5 1 to 5 5). For example, the queue dissipates faster north of NW 81 st St., when the occupancy VSL algorithm is applied compared to the no control case (Figure 5 4). This is also expressed in terms of travel time per vehicle along the freeway stretch from I 195 EB/WB till the exit to 95 th St. The average travel time per vehicle decreases by 0.7 min between 17:45 18:00 pm during the operation of the occupancy based VSL (Figure 5 8). The throughput north of Opa Loc ka Blvd. and north of Biscayne Canal is almost similar between the occupancy based VSL and the no control traffic schemes throughout the whole simulation (Figures

PAGE 94

94 5 6 and 5 7). Moreover, no queues form at the on ramps north of 125 th St. and north of Opa Lo cka Blvd when the occupancy based VSL are active (Figures 5 10 and 5 11). Generally, the occupancy based VSL algorithm outperforms the volume based one at the low demand level, irrespective of the number, location of the signs and the thresholds. The only exception is when two signs are installed upstream of the bottleneck. Then the average network speed is 39.08 mph for the volume based VSL algorithm and 37.95 mph for the occupancy based one (Table 5 2). However, the difference in their performance can be overall considered marginal. Another interesting observation is that the different threshold scenarios did not yield any different results regarding the operation of the algorithms. For instance, the average network speed during the operation of the volume based VSL algorithm is 39.45 mph for both threshold scenarios tested (Table 5 2). The volume based VSL performs better than the occupancy based VSL, during the medium demand scheme. The average network speed is over 37.00 mph when a volume based VSL algor ithm is implemented (Table 5 2). On the contrary it is lower than 37.00 mph while the operation of the occupancy based one (Table 5 2). This occurs irrespective of the number, location of the signs and the selected thresholds. However, their difference in performance is minimal for this demand level too. The different threshold scenarios do not influence the operation of the VSL algorithms at this case as well. When the VSL only case is considered, the maximum average network speed during medium demand is obtained when a volume based VSL algorithm is operating. In this case one sign is located 1.0 mile upstream of the bottleneck source and the

PAGE 95

95 volume thresholds correspond to Scenario 1 (Table 3 3). Then the average network speed is 37.45 mph, which is 2.59 mph higher compared to the no control case (Table 5 2). This improvement of traffic operations on the freeway network is also clearly depicted on the link statistics. The average speed per link is higher across the congested portions of the network over th e whole simulation (Figures 5 12 to 5 16). Throughput north of Opa Locka Boulevard increases by at least 100 veh/hr/lane relative to the no control case at the onset of congestion (i.e. 4 th time period) (Figure 5 17). The increase in throughput is ever bigger north of Biscayne Canal for the same time peri od. It almost reaches u p to 190 veh/hr/lane (Figure 5 18). The average travel time between I 195 and 95 th St. is significantly lesser throu ghout the whole simulation compared to the no control case (Figure 5 20 ). A less significant reduction in travel time is observed between NW 81 st St. and N. Golden Glades Dr. (Figure 5 2 1 ). Finally, the on ramps north of 125 th St. and north of Opa Locka Bl vd remain uncongested for the medium demand case after the implementation of the volume based VSL. It is apparent that the benefits of the VSL installation on the I 95 in Miami compared to the no control case are higher during the medium demand level compa red to the low demand one. During the high demand level, the highest possible average network speed when the volume based VSL are employed is 32.67 mph (Table 5 2). On the contrary, the lowest possible average network speed while the implementation of the occupancy based VSL is 32.84 mph (Table 5 2). Hence, at this demand level the performance of the occupancy based VSL is superior to that of the volume based VSL. This occurs irrespective of the number, location of the VSL signs and the implemented threshol ds. Yet, the difference in their perfo rmance can be considered small.

PAGE 96

96 The maximum average network speed when the occupancy based VSL is tested at the high demand level is 33.18 mph (Table 5 2). This happens when two VSL signs are installed upstream of Bisc ayne Canal and the occupancy thresholds correspond to Scenario 1 (Table 3 1). This is an improvement of 1.2 mph relative to the no control case. This improvement is also illustrated in the charts depicting the average speed per link throughout the network (Figures 5 2 4 to 5 2 8 ). At the end of the simulation, (i.e. 18:15 18:30 pm) the severity of the breakdown between NW 103 rd St. and NW 53 rd St. is smaller compared to the no control case (Figure 5 2 8 ). Throughput increases by 100 veh/hr/lane north of Opa Locka Blvd. and by 30 veh/hr/lane north of Biscayne Canal when congestion breaks out (Figure 5 28 and 5 30 ). The average travel time required per vehicle to cross the two examined routes (i.e. I 195 EB/WB to 95 th St. and 81 st St. to N. Golden Glades Dr.) i s lower relative to the no control case (Figures 5 3 1 and 5 3 2 ). Moreover, vehicles encounter lesser delay at the on ramp north of 125 th St. between 16:30 17:30 pm when the occupancy based VSL is implemented (Figure 5 3 3 ). Generally, the occupancy based VSL control slightly enhances traffic operations on the I 95 in Miami even during the high demand case. When the VSL are implemented during the low and medium demand scenarios, the maximum average network speed (i.e. 40.26 mph) is obtained if the VSL sign is located 1.0 mile upstream of the bottleneck (Table 5 2). However, when demand is high, the best performance (i.e. average network speed equal to 33.18 mph) is achieved when two VSL sings are installed. Thus, the higher the demand the lengthier the porti on of the freeway that is VSL operated should be.

PAGE 97

97 Another interesting remark is that the volume based VSL operates more efficiently only during the medium demand level. However, during the operation of the volume based VSL the speed limits are never decre mented by more than 5.0 mph, since the flow rate in the proximity of the VSL signs never exceeds the maximum volume threshold. Hence, since the volume based VSL is less restrictive, it should be expected that its performance would be superior both for the medium and low demand case s The analysis of the traffic operations on Miami I 95 during the implementation of the VSL has pointed out several important observations regarding their operation and performance. The performance of the VSL is dependent on the type of algorithm applied, nu mber and location of the VSL signs. However, the different thresholds do not affect operations. These observations are similar across all the three examined demand scenarios. The analysis also demonstrated that if demand increases on the network then it i s necessary that a longer stretch of the freeway is VSL operated for the level of service to remain higher. Finally, the deployment of VSL is mostly beneficial during the medium demand scenario If the demand is low or excessively high the deployment of VS L marginally improves traffic operations compared to the no control case. 5.1 .3 Ramp Metering only Case Regarding the ramp metering only case two different ramp metering algorithms were tested as described in Chapter 3; the ALINEA ramp metering algorithm a nd the Fuzzy Logic ramp metering algorithm. The implementation of the ALINEA ramp metering algorithm yields an average network speed of 40.56 mph during the low demand scenario (Table 5 3). This happens when the ALINEA ca libration parameter equals 1.17 ve h/min. When the calibration

PAGE 98

98 parameter is equal to 0.32 speed slightly increases up to 40.70 mph. For both cases the ALINEA ramp metering algorithm performs better compared to the no control case. Wi th respect to throughput, the performance of the ALINEA algorithm is similar to that of the other traffic control strategies (Figures 5 7 and 5 8). However, the average travel time per vehicle between I 195 EB/WB and 95 th St. is 6.0 min for the 9 th time pe riod (Figure 5 8). This is a significant drop compared to the no control case, when the travel time reaches 8.5 min per vehicle for the same time frame. On the contrary, travel time from NW 81 st St. to N. Golden Glades Dr. obtains its maximum value (i.e. 2 0.0 min/veh) during the operation of the ALINEA, since the algorithm regulates traffic inefficiently at the on ramp north of 81 st St (Figure 5 9). Traffic is also restricted at the on ramps north of NW 125 th St. and north of Opa Locka Blvd. Each vehicle en tering the freeway from NW 125 th St. is approximately delayed 0.3 min after 16:45 pm (Figure 5 10). Delay per vehicle hikes up to 5.50 min between 17:45 18:00 pm at the on ramp north of Opa Locka Blvd. (Figure 5 11). During the medium demand scheme, the best performance of the ALINEA ramp metering algorithm is achieved when the ALINEA calibration parameter equals 1.17 veh/min. Then the average network speed is 40.07 mph (Table 5 3). Since, the average network speed for the same demand level during the no control case was 34.86 mph the capacity of the ALINEA algorithm to manage traffic efficiently is clearly realized. Throughput north of Opa Locka Blvd. is 40 veh/hr/lane higher during the operation of the ALINEA algorithm relative to the no control case wh en congestion breaks out (i.e. 16:15 16:45 pm) at the medium demand level (Figure 5 17). The throughput increase

PAGE 99

99 is even bigger north of Biscayne Canal. When the ALINEA algorithm is operating 110 veh/hr/lane more pass through the latter location between 16:15 16:45 pm compared to the no control case (Figure 5 18). The average travel time required for each vehicle to cross the two examined freeway routes is similar between the low and the medium demand scenario when the ALINEA algorithm is implemented. The travel time from I 195 EB/WB to 95 th St. is still 6.0 min/veh between 17:30 17:45 pm (Figure 5 20 ) and 20 min/veh from 81 st St. to N. Golden Glades (Figure 5 2 1 ). However, this occurs at the expense of further spillback on the metered on ramps. The a verage delay per vehicle at the entry from 125 th St. rises from 0.4 min (i.e. low demand scenario) up to 4.0 min (i.e. medium demand scenario) after 16:45 pm (Figure 5 2 2 ). Similarly delay increases by 1.20 min/veh after 17:45 pm at the entry from Opa Lock a Blvd relative to the low demand scenario (Figure 5 2 3 ). While demand is high, the ALINEA algorithm performs once more better when the ALINEA calibration parameter equals 1.17 veh/min. At this case the average network speed is 39.50 mph (Table 5 3). Since the average network speed was 40.07 mph during the medium demand scenario, it is obvious that the ALINEA algorithm succeeds in sustaining a good level of service along the corridor despite the demand increase. This is also depicted in the average link st atistics. Average travel time per vehicle on the tested freeway routes, and delay per vehicle at the two examined on ramps are similar throughout the whole simulation between the medium and the high demand scenarios (Figures 5 3 1 to 5 3 4 ). Throughput incre ases by 30 veh/hr/lane at the onset of congestion (i.e. 16:45) north of Opa Locka Blvd compared to the no control case

PAGE 100

100 (Figure 5 2 9 ). However, it is similar to that of the other control measures for the whole duration of the simulation north of Biscayne Ca nal (Figure 5 30 ). The implementation of the Fuzzy Logic ramp metering algorithm at the low demand level results in an average network speed of 41.94 mph when the fuzzy rule weights correspond to those of Table 3 6 (Table 5 3). However, when the implemente d rule weights correspond to those of Table 3 14 the average network speed decreases by 1.63 mph (Table 5 3). The degraded performance of the algorithm when the modified weights (Table 3 14) are used appears across all the examined demand levels (Table 5 3 ). This occurs since the algorithm becomes less restrictive with the adjusted weights. Hence, traffic enters the mainline network at a higher rate and congestion builds up more rapidly. Although traffic conditions improve at the on ramps at this case, the average network speed drops as the largest proportion of traffic travels on the mainline. The average network speed increases by 1.24 mph when the Fuzzy Logic algorithm is implemented compared to the ALINEA one (Table 5 3). Thus, the operation of the Fuzzy Logic ramp metering algorithm is superior to that of the ALINEA algorithm during the low demand scenario. This superiority is also depicted in the average link statistics. The average travel time from I 195 EB/WB till the exit to 95 th St. is 2.0 min less per vehicle between 17:30 17:45 pm when the Fuzzy Logic algorithm is applied (Figure 5 8). It should be noted that on this route the latter algorithm maint ains the travel time around 3.7 min/veh throughout the simulation time line. T his value is very close to the free flow speed travel time corresponding to this path (i.e. 3.36 min/veh). But on the other route also, the travel time per vehicle is 6.0 min less between 17:45 18:00 relative to the ALINEA algorithm (Figure 5 9). The del ay per vehicle at the on ramp

PAGE 101

101 north of 125 th St. is slightly higher when the Fuzzy Logic ramp metering algorithm is implemented (Figure 5 10). On the other hand, at the entry from Opa Locka Blvd. delay is approximately 1.0 min/veh higher during the congest ed time periods when the ALINEA algorithm is operating (Figure 5 11). An interesting remark considering the operation of the Fuzzy Logic algorithm is that it manages to spread congestion on the ramps more evenly during the simulation compared to the ALINEA algorithm (Figures 5 10 and 5 11). During the medium demand scenario the operation of the Fuzzy Logic ramp metering algorithm results in an aver age network speed of 40.21 mph. Throughput north of Opa Locka Blvd. increases by 50 veh/hr/lane compared to th e no control case when congestion begins (i.e. 4 th time period) (Figure 5 17). The increase in throughput is even higher (100 veh/hr/lane) north of Biscayne Canal (Figure 5 18). The travel time per vehicle on both selected routes is exactly the same compar ed to the low demand scenario (Figures 5 20 and 5 2 1 ). The algorithm prevents the deterioration of traffic conditions on the mainline network despite the demand increase. However, congestion at the on ramps is worsened. The delay vehicles experience at the entry from NW 125 th St. range between 0.25 min/veh and 1.25 min/veh throughout the simulation for the low demand case. Now the delay per vehicle ranges between 1.00 3.25 min (Figure 5 2 2 ). Accordingly, there is an average delay increase of 0.5 min/veh f or traffic entering from Opa Locka Blvd during the medium demand scenario. The average network speed when the Fuzzy Logic ramp metering algorithm is implemented at the high demand level is 40.77 mph (Table 5 3). It is clear that traffic conditions on the mainline remain unchanged between the medium and the high

PAGE 102

102 demand scenarios during the operation of the Fuzzy Logic algorithm. Throughput north of Opa Locka Blvd. and north of Biscayne Canal is similar at both demand cases (Figures 5 2 9 and 5 30 ). The same also occurs regarding the travel time per vehicle on the two examined routes of t he freeway network (Figures 5 31 and 5 3 2 ). The only observable difference is that of congestion at the on ramps. Delay per vehicle slightly increases at both on ramps (i.e. north of NW 125 th St. and north of Opa Locka Blvd) after the increase in demand. The results of this study indicate that the performance of the ALINEA ramp metering algorithm slightly changes for the different tested values of the ALINEA calibration parame ter K R On the contrary, field implementations of the algorithm have shown that a value of 1.17 veh/min produces good results with respect to its performance (Papageorgiou et al., 1997). The performance of the Fuzzy Logic ramp metering algorithm varies mor e significantly according to the implemented fuzzy rule weights. This analysis demonstrated that the algorithm performs better when the rule weights proposed by Meldrum and Taylor (2000) are implemented irresp ective of the demand level. A common characteri stic of both algorithms is that they manage to regulate incoming traffic sufficiently enough, so that traffic conditions on the mainline network are not hugely deteriorated. They both achieve this even at the very high demand level. However, their implemen tation induces spillback on the ramps where queues grow as demand grows. The difference between the two algorithms is that throughout the operation of the Fuzzy Logic ramp metering delays on the on ramps are lower and more evenly spread across the differen t time periods.

PAGE 103

103 Generally, the implementation of the Fuzzy Logic ramp metering algorithm results in a higher level of service along the freeway compared to that of the ALINEA one. This observation is common among all the examined demand levels. This happen s mainly for two reasons. The first one is that the Fuzzy Logic ramp metering algorithm uses more congestion indicators to determine the metering rate. This means that the algorithm is fed with traffic counts collected downstream, upstream and on the respe ctive on ramp, while the ALINEA algorithm utilizes only downstream data to calculate the metering rate. Thus, the Fuzzy Logic algorithm captures more accurately the effects of congestion on the network and is capable of responding more efficiently. The sec ond reason is that the Fuzzy Logic algorithm is a proactive algorithm rather than a reactive one, as the ALINEA is. During the operation of the Fuzzy Logic algorithm the metering rates are adjusted continuously according to the prevailing traffic condition s. On the contrary, the ALINEA algorithm restricts traffic only when congestion ha s been detected on the network. It is also apparent that both ramp metering algorithms outperform the implemented VSL ones. This difference is more profound at the medium and the high demand levels. It can be partially ascribed to the fact that the VSL are implemented only on a small portion of the freeway, while the ramp meters are placed at all the on ramps which are located along the congested part of the freeway. 5.1 .4 Int egrated Control Case The ALINEA and the Fuzzy Logic ramp metering algorithms were combined with the occupancy and volume based VSL algorithms so that their integrated operation would be evaluated. All the simulated scenarios corresponding to the integrated control case appear in detail in Chapter 3 (Tables 3 10 to 3 13).

PAGE 104

104 The concurrent implementation of the ALINEA ramp metering algorithm with the volume based VSL one yields an average network speed of 41.38 mph during the low demand case. This occurs for t he installation of one VSL sign 0.5 miles upstream of the bottleneck source, volume thresholds corresponding to Scenario 1 (Table 3 3) and an ALINEA calibration parameter of 1.17 veh/min (Table 5 4). If the ALINEA calibration parameter decreases to 0.32 v eh/min then the average network speed becomes 41.10 mph. In the case that just the VSL sign is moved 1.0 mile upstream of the bottleneck, then it reaches 41.24 mph (Table 5 4). It is obvious that for different settings of the parameters determining the ope ration of each algorithm the average network performance is slightly changed. A careful examination of Table 5 3 reveals that this happens across all the demand levels when the ALINEA algorithm is combined with the volume based VSL one. The network wide st atistics corresponding to the joint operation of the ALINEA ramp metering algorithm and the occupancy based VSL at the three examined demand levels are presented in Table 5 5. A comparison of the results in Tables 5 4 and 5 5 exhibits the similarity of the performance of the ALINEA algorithm either it is combined with occupancy based VSL one or with the volume based. The similarity in performance is bol d across all the demand levels. The simultaneous operation of ALINEA and VSL increase significantly the av erage speed per link on the I 95 in Miami during the low demand conditions relative to the no control case (Figures 5 1 to 5 5). This reduction of the severity, length and duration of the breakdowns on the network is more substantial during the medium and the high demand schemes (Figures 5 12 to 5 16 and 5 2 4 to 5 2 8 ). Throughput north of Opa

PAGE 105

105 Locka Blvd. and north of Biscayne Canal is similar between the ALINEA VSL case compared to the ALINEA only case (Figures 5 6 and 5 7). The travel time per vehicle from I 195 EB/WB to 95 th St. and from NW 81 st St. to N. Golden Glades is also the same between these two types of traffic control s chemes (Figures 5 20 and 5 2 1 ). The similar performance of the ALINEA VSL integrated control and the ALINEA only case with respect to throughput and travel time occurs irrespective of the examined demand scenarios. Thus, both the abovementioned traffic management schemes sustain similar traffic conditions on the mainline network on the I 95 in Miami. However, traffic on the on ramps is more efficiently regulated when the combination of ALINEA and VSL is implemented. For the low demand case the average delay per vehicle is reduced both at the entry form NW 125 th St. and at the entry from Opa Locka Blvd. (Figures 5 10 and 5 11). For the other two demand cases this reduction is only significant at the on ramp north of NW 125 th St. (Figures 5 2 2 and 5 3 3 ). The average network statistics corresponding to the combined operation of the Fuzzy Logic ramp metering algorithm with the volum e based VSL and the occupancy based VSL are shown in Tables 5 6 and 5 7 respectively. A comparison of these results indicates that the average network speed is higher across all the demand scenarios when the Fuzzy Logic algorithm is combined with the occup ancy based VSL compared to the case that it is operating together with the volume based. Moreover, it appears that the number and location of the VSL signs, as well as the occupancy or volume thresholds do not impact substantially the performance of the Fu zzy ramp metering and the VSL at a specific demand level. However, the adjusted fuzzy rule weights (Table 3

PAGE 106

106 14) clearly degrade the performance of the integrated control, ir respective of the demand level. The joint operation of Fuzzy Logic ramp metering an d VSL minimize the effects of congestion on the I 95 in Miami (Figures 5 1 to 5 5, 5 12 to 5 16 and 5 2 4 to 5 2 8 ). It is noticeable that congestion is totally mitigated at the end of the simulation (i.e. 18:15 18:30 pm) for all of the three examined deman d levels. A comprehensive observation of the average link statistics (i.e. throughput, average travel time per route, delay per vehicle at the on ramps) illustrates the equivalent performance of the combination of Fuzzy Logic ramp metering and VSL with the Fuzzy Logic ramp metering only case. Although there might exist minor changes in the link statistics between these traffic control strategies for the low and medium demand scenarios, the results are completely identica l for the high demand case. So far th e results pertaining to throughput have been presented for two specific locations upstream of the breakdown point; North of Opa locka Blvd. and North of Biscayne Canal. However, the assessment of throughput should be also conducted downstream of the bottle neck source to determine the effects of each traffic management scheme on traffic operations. Thus, throughput has been also obtained South of US 441 for the best implementation of each control case during the medium demand scenario. The pattern of through put South of US 441 for the different control cases is similar to that of the locations upstream of the breakdown. In the beginning of congestion (i.e. 4 th time period) throughput is higher relative to the no control case when VSL, ramp metering or their c ombination is implemented (Figure 5 19 ). The pattern of

PAGE 107

10 7 throughput though is similar among the latter three traffic control cases over the entire duration of the simulat ion (Figure 5 19 ). According to the results of this study the combination of the Fuzzy Logic ramp metering and the VSL manages traffic more efficiently compared to the combined operation of the ALINEA ramp metering algorithm and the VSL. The difference in the performance of the two combinations is not significant though. However, a careful e xamination of the average network statistics clearly indicates that the average network speed is about 1.00 mph higher during the concurrent operation of the Fuzzy Logic ramp metering algorithm and the VSL (Tables 5 4 to 5 7) relative to the other combinat ion. This observation is uniform across all the three tested demand levels. It is also clearly depicted on the average speed per link charts that congestion is fully resolved when the Fuzzy Logic ramp metering is implemented along with the VSL by the end o f the simulation time line. This does not occur when the ALINEA ramp metering algo rithm is combined with the VSL. It must be noted that in past simulation studies the performance of the integrated control scheme has been found to be substantially better co mpared to the metering only strategy (Heg yi et al., 2004, Papamichail et al., 2008, Ghods et al. 2009, Carlson et al., 2010). The simulation experiments conducted in this analysis have shown that the integrated control performs slightly better relative to the metering only case. This result can be ascribed to the fact that the portion of the network that is VSL operated in this study is rather small compared to its whole extent. In the other studies the simulated networks were small and VSL operated along t heir entire length. However, it is still

PAGE 108

108 verified that congestion is mitigated more efficiently under the integrated control strategy. 5.1 .5 Statistical Analysis of the Simulation Results The simulated traffic management strategies have been compared with respect to the average network speed. Since each scenario has been simulated 10 times the average network speed has been estimated as the mean speed value of the 10 runs. The results have shown that the integrated control outperforms the rest of the contro l cases. Moreover, the implementation of ramp metering yields improved traffic conditions relative to the VSL only case and the no control case. Finally, the average network speed during the operation of the VSL is higher compared to the no control case. H owever, it is essential that the credibility of these results is verified through the use of statistical testing. The average network speeds corresponding to each traffic control strategy are compared to each other to determine whether the observed differences are statistically significant. The average network speeds used in these tests correspond to the best implementation of each control case during the highest demand scenario. The random seeds used for all scenarios are the same Thus, the charac teristics of the traffic stream are similar among the different scenarios and the four examined sample size is 10 (i.e. 10 runs per simulated scenario) the degree of freedoms are 9. The null hypothesis is that the difference between the average network speeds of two traffic control schemes is lower than zero (i.e. H 0 : 1 2 < 0) The alternative hypothesis is that the difference is higher (i.e. H : 1 2 > 0)

PAGE 109

109 iven for each test form the subsequent formula: (5 1) Based on t he statistical tests we can conclude the following at a 95% confidence level (Table 5 8 ): The average network speed during the implementation of the integrated control scheme is higher compared to the rest of the control cases. The average network speed while ramp metering is applied is higher relative to the VSL only case and the no control case. The average network speed during the implementation of the VSL is higher compared to the no control case. Thus, all the initial conclusions have been verified through rigorous statistical analysis. 5.2 Interactions between Ramp Metering and VSL It has been already mentioned that the interactions between ramp metering and VSL were studied by examining the VSL rates along with the metering rates imposed at the on ramps which are located in the proximity of the VSL sings. Three different configurations were simulated with respect to the location of the VSL signs. Initially, one VSL sign was placed immediately downstream of the entrance from Opa Locka Boulevard (Figure 5 3 5 ). Then, the sign was moved just upstream of the same entrance (Figure 5 3 6 ). Finally the installation of two VSL signs was tested with respect to the perfo rmance of traffic on the network. The first was placed upstream of the entrance from Opa Locka Boulevard and the second upstream of the exit to NW 125 th Street (Figure 5 3 7 ). Irrespective of the location of the signs the displayed speed limits were

PAGE 110

110 determi ned according to the traffic counts measured from the loop detectors installed upstream of the exit to the Turnpike. These VSL rates were compared to the estimated metering rates imposed at the on ramp north of Opa Locka Boulevard for every scenario involv ing the concurrent implementation of a VSL and a ramp metering strategy. The comparison of the VSL and metering rates was conducted for each demand scenario separately. Finally, observations pertaining to the interactions between ramp metering and VSL that were identified to be common among many different scenarios are summarized. 5.2 .1 Low Demand Scenario Fuzzy Logic Ramp Metering & Occupancy based VSL The installation of the VSL sign downstream of the metered ramp results in an increase of the VSL rate w hen the metering rate significantly decreases (Figure 5 3 8 ). However, it should be mentioned that the decrease of the metering rate occurs after the VSL rate has attained its minimum value from the beginning of the simulation. This means that as demand is growing the VSL will initially begin to restrict traffic. But when the traffic conditions around the on ramp become congested then the meters significantly decrease the entering traffic leading in an increase of the VSL rate during the subsequent time step s. Considering the same VSL configuration (i.e. number and location of signs), but accounting for fuzzy rule weights which provide priority to on ramp traffic there is no clear interaction between the two systems (Figure 5 3 9 ). The only clear observation i s that the metering rate becomes too oscillatory throughout the most congested time periods. When one or two VSL signs are installed upstream of the examined metered

PAGE 111

111 ramp then the changes in the metering rate are so marginal that again no interactions can be discerned between ramp metering a nd VSL (Figures 5 40 and 5 4 1 ). Fuzzy Logic Ramp Metering & Volume based VSL The interactions between ramp metering and VSL are rather ambiguous when the VSL sign is installed downstream of the metered ramp (Figure 5 4 2 ). At the end of the 6 th time period an abrupt decrease of both the VSL and the metering rates occurs simultaneously. However, at the end of the 8 th time period the decrease of the metering rate happens along with an increase of the VSL one. It should be a lso mentioned that the operation of the VSL system is unstable in the beginning of the simulation. Although the VSL operation becomes stable when the fuzzy rule weights are modified to favor on ramp traffic, the metering rates are then rendered too oscilla tory during the most congested time periods (Figure 5 4 3 ). The operation of the VSL system is complementary to the metering one when the VSL signs (i.e. one or two signs) are set upstream of the metered ramp (Figures 5 4 4 and 5 4 5 ). At this case the patter n of the VSL rates is always in accordance with that of the metering rates. Increases in the metering rate cause increases in the VSL rate and vice versa. Moreover, the intensity of fluctuation of the metering rates follows that of the VSL rates throughout the whole simulation. ALINEA Ramp Metering & Occupancy based VSL The metering rate remains unchanged during the simulation when the VSL sign is located downstream of the examined metered ramp and the ALINEA regulator parameter is equal to 1.17 veh/min (Figure 5 4 6 ). Thus, it is obvious that no interactions between ramp metering and VSL exist in this case. On the contrary, if the regulator parameter decreases to 0.32 veh/min then changes in the metering rates result in changes in the VSL rates (Figure 5 4 7 ).

PAGE 112

112 However, these changes are not uniform throughout the simulation. In the beginning of the 7 th time period a significant decrease in the metering rate results in an increase of the VSL rate. But the effect of a similar reduction of the metering rate at the end of the 9 th time period is exactly the opposite with respect to the VSL rate. The installation of the VSL sign upstream of the metered ramp establishes a reactive relationship between ramp metering and VSL (Figure 5 4 8 ). When the one system becomes restrictive the other one is rendered unrestrictive and vice versa. Finally, the placement of two signs instead of one upstream of the examined metered ramp results in no interactions between the two systems (Figure 5 4 9 ). In this case it is clear that ch anges in the metering rates occur after a long period of time has elapsed since the last change in the VSL rate. ALINEA Ramp Metering & Volume based VSL The interactions between ramp metering and VSL are unclear when the VSL sign is placed downstream of t he entry from Opa Locka Boulevard (Figure 5 50 ). The metering rate decreases to the minimum possible value (i.e. 5 veh/min) in the beginning of the 10 th time period and returns back to its maximum value (i.e. 28 veh/min) at the end of it. As this is the on ly change that occurs to the metering rate throughout the simulation it is obvious that the operation of the two systems in not interrelated. However, if the ALINEA calibration parameter decreases to 0.32 veh/min for the same VSL configuration, it becomes clear that the reduction of the metering rate renders the VSL rate less restrictive temporarily (Figure 5 5 1 ). But after a few minutes the VSL drops down to 45 mph again. The relocation of the VSL sign upstream of the metered ramp makes the operation of th e one system irresponsive to the changes in the operation of the other (Figure 5 5 2 ). During the first

PAGE 113

113 six time periods the VSL rate is rather oscillatory, but this operation induces no changes to the metering rate. On the other hand, changes to the meteri ng rate occur when the operation of the VSL system has stabilized throughout the rest of the simulation. Finally, the placement of two VSL signs upstream of the examined metered ramp results in a complementary operation of the VSL and ramp metering (Figure 5 5 3 ). A decrease in the metering rate is followed by a decrease in the VSL rate and vice versa. 5.2 .2 Medium Demand Scenario Fuzzy Logic Ramp Metering & Occupancy based VSL Three main observations can be made with respect to the case that one VSL sign i s installed downstream of the entry from Opa Locka Boulevard. In the beginning of the simulation the VSL rate drops as demand is increasing. However, during that period the metering rate remains unaltered (Figure 5 5 4 ). When the VSL rate increases in the m iddle of the sixth time period the metering rate becomes more restrictive. Due to growing demand the VSL rate will attain its minimum value (i.e. 40mph) after 25 minutes but this change does not affect the operation of the metering system. No interactions exist between ramp metering and VSL when the fuzzy rule weights are modified to favor on ramp traffic and the VSL configuration remains unchanged (Figure 5 5 5 ). Despite the changes of the VSL rate the metering rate stays stable over the whole simulation. T he interactions between ramp metering and VSL become unclear when the VSL sign is moved upstream of the metered ramp (Figure 5 5 6 ). Although both systems behave similarly at the onset of congestion (i.e. 4 th time period), the metering rate stabilizes at 10 veh/min after forty minutes and does not fluctuate in spite of the changes that occur

PAGE 114

114 in the VSL rate. Similar conditions prevail if two VSL signs are installed upstream of the metered ramp instead of one (Figure 5 5 7 ). Fuzzy Logic Ramp Metering & Volume based VSL There are no interactions between ramp metering and VSL when the VSL sign is installed downstream of the metered ramp. In the case that the fuzzy rule weights attain their initial values the VSL rate remains almost unchanged at 45 mph throughou t the whole simulation (Figure 5 5 8 ). On the other hand, when the fuzzy rule weights are modified to provide priority to the on ramp traffic it is the ramp metering rate that stays invariable (Figure 5 5 9 ). The operation of the two systems becomes compleme ntary though when the VSL sign is located upstream of the examined metered ramp. Then the metering rate increases when the VSL rate increases and vice versa (Figure 5 60 ). This pattern is not sustained when two VSL sings are placed upstream of the metered ramp. In this case the operation of the one system does not impact that of the other (Figure 5 6 1 ). ALINEA Ramp Metering & Occupancy based VSL The results with respect to the interactions between ramp metering and VSL are incoherent when the VSL sign is l ocated downstream of the metered ramp and the ALINEA calibration parameter equals 1.17 veh/min (Figure 5 6 2 ). During the initial five time periods several changes in the VSL rate do not incur any changes to the metering rate. On the contrary, throughout th e rest of the simulation reductions in the VSL rate cause either increase or decrease in the metering rate. Thus, no definitive answer can be given about the interrelationship of the two systems. If the ALINEA calibration parameter is reduced to 0.32 veh/m in and the VSL configuration remains unchanged then the interactions become more consistent. More restrictive metering rates drive the VSL system to less restrictive

PAGE 115

115 operation (Figure 5 6 3 ). However, this change occurs after a significant amount of time ha s elapsed. The installation of the VSL sign upstream of the metered ramp yields no conclusive answer regarding the interactions of the two traffic control strategies. The operation of the one system does not affect that of the other (Figure 5 6 4 ). On the c ontrary, the reduction of the metering rate forces the VSL rate to increase when two VSL signs are installed upstream of the examined metered on ramp (Figure 5 6 5 ). ALINEA Ramp Metering & Volume based VSL The interactions between ramp metering and VSL are not transparent when the VSL sign is located upstream of the metered ramp and the ALINEA calibration parameter is equal to 1.17 veh/min. During the beginning of the simulation changes in the VSL rate do not incur any changes in the ramp metering rate (Fig ure 5 6 6 ). However, after the fifth time period a reduction in the metering rate can either increase or decrease the VSL rate. On the other hand, no potential interactions can be identified when the calibration parameter is reduced to 0.32 veh/min and the VSL configuration remains the same. At this case the metering rate maintains its maximum value (i.e. 29 veh/min) for the first five time periods and then it suddenly drops to its minimum value (i.e. 5 veh/min) till the middle of the tenth time period (Figu re 5 6 7 ). The changes that occur in the VSL rate do not affect the signs are placed upstream of the examined metered ramp (Figure 5 6 8 and 5 6 9 ). 5.2 .3 High Demand Scen ario Fuzzy Logic Ramp Metering & Occupancy based VSL The VSL rate decreases to its minimum value (i.e. 40 mph) in the beginning of the 4 th time period when the VSL sign is installed downstream of the entry from Opa Locka Boulevard (Figure 5 70 ). This

PAGE 116

116 drop does not inflict any changes in the metering rate though. On the contrary, when the metering rate decreases by 5 veh/min later in the simulation the VSL rate increases to 45 mph (Figure 5 70 ). Early changes in the VSL rate do not affect the operation of t he metering system because demand is low at the same time. However, the growth of traffic demand in the next time periods creates interactions between the two systems. In this specific case a restrictive operation of the meters causes an unrestrictive oper ation by the VSL. For the same VSL configuration but for modified fuzzy rule weights no interactions exist between ramp metering and VSL (Figure 5 7 1 ). The metering rate remains nearly invariable throughout t he whole simulation time line. An interesting ob servation though is that both the metering rate and the VSL rate are less restrictive compared to the preceding case (Figure 5 7 1 ). The operation of the two systems becomes complementary when the VSL sign is moved upstream of the entrance from Opa Locka Bo ulevard. Both the metering and the VSL rate decrease in the middle of the fifth time period and despite some oscillations of the VSL rate follow the same pattern till the end of the simulation (Figure 5 72). Finally, the interactions between ramp metering and VSL are unclear when two VSL signs are installed upstream of the examined metered ramp. An abrupt decrease of the metering rate in the beginning of the 7 th time period initially increases the VSL rate (Figure 5 73). However, VSL become more restrictive subsequently without affecting the operation of the meter though. It is important to be noticed though that when two VSL signs are installed upstream of the metered ramp the initial metering rate remains unchanged for a larger amount of time compared to t he rest of the VSL configurations. This might be due to the fact that the

PAGE 117

117 VSL regulate traffic earlier in the network, thus preventing downstream congestion to impact traffic conditions around the metered ramp faster. Fuzzy Logic Ramp Metering & Volume bas ed VSL The VSL rate remains 45 mph during most of the simulation duration when the VSL sign is located downstream of the examined metered on ramp (Figure 5 74). Yet, during the first five time periods it is more oscillatory compared to the rest of the sim ulation. The oscillatory performance of the VSL in the beginning of the simulation coincides with a lesser restrictive operation of the ramp metering (Figure 5 74). After the metering system starts to operate in a more restrictive manner the VSL rate stays steady at 45 mph. It is clear that as congestion begins to spread in the network both the metering and the VSL rates decrease. The modification of the fuzzy rule weights results in no interactions between ramp metering and VSL for the abovementioned VSL c onfiguration. The ramp metering rate maintains a high value (i.e. around 15 veh/min) and does not vary over the simulation time line (Figure 5 75). Therefore, it would be expected that the VSL would attain the smallest possible rate. On the contrary, the V SL rate becomes more oscillatory compared to the previous case. The operation of the VSL complements that of the ramp metering when the VSL sign is located upstream of the entry ramp from Opa Locka Blvd (Figure 5 76). The VSL rate and the metering rate fol low almost the same pattern throughout the whole simulation. The installation of two VSL signs upstream of the examined metered ramp forces the metering system to become more restrictive later in the simulation (Figure 5 7 7 ). Moreover, the VSL rate is more oscillatory in the beginning of the simulation when the metering rate is rather high.

PAGE 118

118 ALINEA Ramp Metering & Occupancy based VSL The reduction of the metering rate to its minimum value (i.e. 5 veh/min) forces the VSL rate to increase in the subsequent time periods (Figure 5 7 8 ). It is also clear that in the beginning of the simulation the VSL starts to restrict traffic before the metering operation becomes restrictive. This occurs because congestion grows initially around the area that the detectors whi ch regulate the VSL are placed (Figure 5 3 5 ). These observations are made when the VSL sign is installed downstream of the examined metered ramp and the ALINEA calibration parameter is equal to 1.17 veh/min. When the value of the calibration parameter drop s to 0.32 veh/min changes in the metering rate become smoother. Thus it takes more time for changes in the metering rate to affect the VSL operation (Figure 5 79 ). Besides that, the interactions between ramp metering and VSL remain the same when the value of the calibration parameter decreases. When the VSL sign is moved upstream of the metered on ramp then both the VSL and the metering rate follow the same pattern (Figure 5 8 0 ). Thus, the one system complements the other in the effort to regulate traffic a nd mitigate congestion. The results with respect to the interactions between ramp metering and VSL are rather ambiguous when two VSL signs are installed upstream of the metered on ramp. In this case a decrement of the VSL rate can either cause an increase or decrease of the metering rate (Figure 5 8 1 ). ALINEA Ramp Metering & Volume b ased VSL The VSL rate remains steady at 45 mph throughout most of the simulation when the VSL sign is located downstream of the entry from Opa Locka Blvd and the ALINEA calibra tion parameter is 1.17 veh/min (Figure 5 8 2 ).This outcome is expected since demand is high but the throughput never exceeds the upper limit set to decrement the VSL rate to its lowest possible value (i.e.

PAGE 119

119 40 mph). On the contrary, the metering rate decreas es to its minimum value at the end of the 6th time period and oscillates around this va lue till the beginning of the 11 th time period. However, this change does not affect the operation of the VSL (Figure 5 83). The reduction of the calibration parameter t o 0.32 veh/min does not alter the performance of the VSL. It simply renders the transition of the ALINEA algorithm from restrictive to unrestrictive operation more tardy (Figure 5 84). Therefore, no major change on the VSL and metering rate patterns occurs This latter remark is also valid for the case that one or two VSL signs are installed upstream of the examined metered ramp. The only small difference now is that the VSL rate becomes slightly more oscillatory, especially at the later and more congested time periods of the simulation (Figure 5 85). 5.2 .4 Summary of the Interactions between Ramp Metering and VSL The detailed description of the interactions between ramp metering and VSL for all the possible combinations of algorithms during all the tested d emand levels reveals a general rule that dictates the concurrent operation of the two systems. It is noted that in the beginning of each simulation traffic is initially restricted by the operation of the VSL. This phenomenon occurs because traffic conditio ns are initially deteriorated around the area that the detectors determining the operation of the VSL are installed. Thus, it is the VSL rate that is firstly decreased due to the onset of congestion. The operation of the metering system is influenced after congestion spreads upstream to the proximity of the on ramp north of Opa Locka Blvd. At this point the metering rate becomes more restrictive in order to improve traffic flow around the ramp. But the improvement of the traffic flow around the ramp is grad ually migrating downstream back to the location of the detectors that are related to the VSL operation. Hence, it is the VSL operation that

PAGE 120

120 now becomes less restrictive. This pattern regarding the concurrent operation of the two systems is almost uniform f or all the combinations of ramp metering and VSL algorithms. However, it must be mentioned that during the latest time periods of the simulations, when congestion is very severe along the whole network, both the metering and the VSL rate attain their most restricted values in order to regulate effectively traffic. The response of the ramp metering system to the spread out of congestion depends upon the location of the VSL signs. The furthest upstream from the on ramp the VSL signs are installed the later th e reduction of the metering rate will occur throughout the simulation. For example, the installation of two VSL signs upstream of the examined metered on ramp restricts traffic earlier along the network. Therefore, it takes more time before congestion expa nds to affect traffic operations around the on ramp. When one sign is installed upstream of the ramp then the response of the ramp metering algorithm occurs earlier relative to the latter case. On the other hand, the change of the VSL operation in respons e to the reduction of the metering rate is based upon the settings of the ramp metering algorithm. In the case the ALINEA ramp metering algorithm is combined with the VSL, the ALINEA calibration parameter determines this interaction. Since lower values of the parameter induce slower transition from unrestrictive to restrictive metering rates, the VSL rate increases later in the simulation when the ALINEA calibration parameter has a low value. However, a similar operation does not exist when the Fuzzy Logic ramp metering algorithm is combined with the VSL. In this case the modification of the fuzzy rule weights does not result in any interactions between ramp metering and VSL. This occurs because the weights are modified so that traffic on the ramps is not si gnificantly

PAGE 121

121 restricted. Thus, the metering rate slightly oscillates around its maximum value over the whole simulation. Finally, the alternative volume and occupancy thresholds that were examined with respect to the VSL operation did not alter the interact ions between ramp metering and VSL relative to the base cases. Another notable observation is that the VSL rate never drops to the minimum possible value (i.e. 40 mph) when the volume based VSL is implemented concurrently with a ramp metering algorithm. T his occurs because the upper volume limit set for decrementing the VSL rate to 40 mph is never exceeded. Thus, the VSL rate oscillates between 45mph and 50mph during the simulation. As a result the operation of the metering system is now more restrictive c ompared to the combination of the occupancy based VSL and ramp metering. Moreover, the VSL rate becomes more oscillatory throughout the whole simulation. The unrestrictive and unstable operation of the volume based VSL when combined with ramp metering resu lts in a downgraded network performance relative to the combination of occupancy based VSL and ramp metering. This has been already demonstrated earlier in this chapter when the results pertinent to the integrated control case were analyzed.

PAGE 122

122 T able 5 1. Cumulative network statistics on the I 95 in Miami under no traffic management scheme. Scenario # Demand (veh/hr) Travel Distance Total (miles) Total Travel Time (hours) Average Network Speed (mph) 1 Low 297095.70 7587.60 39.16 2 Medium 299185.80 8583.69 34.86 3 High 295597.80 9244.23 31.98

PAGE 123

123 Table 5 2. Cumulative network statistics on the I 95 in Miami during the operation of the tested VSL algorithms. Scenario # Demand (veh/hr) VSL Algorithm Number of VSL signs Locations of VSL signs VSL operation Travel Distance Total (miles) Total Travel Time (Hours) Average Network Speed (mph) 4 A Low Volume One 0.5 miles upstream Moderate 296658.10 7520.08 39.45 B Low Volume One 0.5 miles upstream Aggressive 296658.10 7520.08 39.45 C Low Volume One 1.0 mile upstream Moderate 297557.80 7567.37 39.32 D Low Volume Two 1.0 mile upstream Moderate 297217.50 7605.71 39.08 E Low Occupancy One 0.5 miles upstream Moderate 297249.80 7412.90 40.10 F Low Occupancy One 0.5 miles upstream Aggressive 297249.80 7412.90 40.10 G Low Occupancy One 1.0 mile upstream Moderate 297693.40 7394.50 40.26 H Low Occupancy Two 1.0 mile upstream Moderate 297452.20 7837.61 37.95 5 A Medium Volume One 0.5 miles upstream Moderate 299073.50 8068.72 37.07 B Medium Volume One 0.5 miles upstream Aggressive 299073.50 8068.72 37.07 C Medium Volume One 1.0 mile upstream Moderate 299300.60 7992.01 37.45 D Medium Volume Two 1.0 mile upstream Moderate 299190.40 8047.80 37.18 E Medium Occupancy One 0.5 miles upstream Moderate 299016.40 8174.78 36.58 F Medium Occupanc y One 0.5 miles upstream Aggressive 299016.40 8174.78 36.58 G Medium Occupancy One 1.0 mile upstream Moderate 299003.70 8211.54 36.41 H Medium Occupancy Two 1.0 mile upstream Moderate 299775.00 8193.34 36.59 6 A High Volume One 0.5 miles upstream Moderate 297006.80 9090.57 32.67 B High Volume One 0.5 miles upstream Aggressive 297006.80 9090.57 32.67 C High Volume One 1.0 mile upstream Moderate 297425.40 9181.27 32.40 D High Volume Two 1.0 mile upstream Moderate 296229.90 9308.40 31.82 E High Occupancy One 0.5 miles upstream Moderate 298077.70 9076.30 32.84 F High Occupancy One 0.5 miles upstream Aggressive 298077.70 9076.30 32.84 G High Occupancy One 1.0 mile upstream Moderate 297703.90 8981.74 33.15 H High Occupancy Two 1.0 mile upstream Moderate 297927.10 8979.21 33.18

PAGE 124

124 Table 5 3. Cumulative network statistics on the I 95 in Miami during the operation of the tested Ramp Metering algorithms. Scenario # Demand (veh/ hr) Ramp Metering Algorithm ALINEA Calibration Parameter (veh/min) FUZZY Rule Weights Travel Distance Total (miles) Total Travel Time (hours) Average Network Speed (mph) 7 A Low ALINEA 1.17 296385.40 7308.24 40.56 B Low ALINEA 0.32 296093.80 7274.60 40.70 C Low Fuzzy Logic Proposed 296498.90 7069.88 41.94 D Low Fuzzy Logic Modified 297513.10 7380.29 40.31 8 A Medium ALINEA 1.17 297914.40 7435.14 40.07 B Medium ALINEA 0.32 297838.80 7536.72 39.52 C Medium Fuzzy Logic Proposed 297426.10 7396.89 40.21 D Medium Fuzzy Logic Modified 297847.20 7628.05 39.05 9 A High ALINEA 1.17 298303.70 7552.94 39.50 B High ALINEA 0.32 298480.90 7621.61 39.16 C High Fuzzy Logic Proposed 298118.30 7312.49 39.83 D High Fuzzy Logic Modified 298120.80 7762.89 38.40

PAGE 125

125 Table 5 4. Cumulative network statistics on the I 95 in Miami during the concurrent operation of the ALINEA ramp metering and the volume based VSL algorithm. Scenario # Demand (veh/hr) ALINEA Calibration Parameter (veh /min) Number of VSL signs Locations of VSL signs VSL operation Travel Distance Total (miles) Total Travel Time (Hours) Average Network Speed (mph) 10 A Low 1.17 One 0.5 miles upstream Moderate 296654.30 7168.64 41.38 B Low 0.32 One 0.5 miles upstream Moderate 296817.00 7221.47 41.10 C Low 1.17 One 0.5 miles upstream Aggressive 296654.30 7168.64 41.38 D Low 1.17 One 1.0 mile upstream Moderate 296592.20 7192.58 41.24 E Low 1.17 Two 1.0 mile upstream Moderate 297563.90 7319.28 40.66 11 A Medium 1.17 One 0.5 miles upstream Moderate 298469.00 7376.31 40.46 B Medium 0.32 One 0.5 miles upstream Moderate 297753.50 7465.89 39.88 C Medium 1.17 One 0.5 miles upstream Aggressive 298469.00 7376.31 40.46 D Medium 1.17 One 1.0 mile upstream Moderate 297844.20 7365.79 40.44 E Medium 1.17 Two 1.0 mile upstream Moderate 297995.20 7493.28 39.77 12 A High 1.17 One 0.5 miles upstream Moderate 298592.80 7576.39 39.41 B High 0.32 One 0.5 miles upstream Moderate 298346.90 7560.03 39.46 C High 1.17 One 0.5 miles upstream Aggressive 298592.80 7576.39 39.41 D High 1.17 One 1.0 mile upstream Moderate 298455.40 7680.12 38.86 E High 1.17 Two 1.0 mile upstream Moderate 297932.10 7649.51 38.95

PAGE 126

126 Table 5 5. Cumulative network statistics on the I 95 in Miami during the concurrent operation of the ALINEA ramp metering and the occupancy based VSL algorithm. Scenario # Demand (veh/hr) ALINEA Calibration Parameter (veh/min) Number of VSL signs Locations of VSL s igns VSL operation Travel Distance Total (miles) Total Travel Time (Hours) Average Network Speed (mph) 13 A Low 1.17 One 0.5 miles upstream Moderate 296700.40 7159.16 41.44 B Low 0.32 One 0.5 miles upstream Moderate 296271.90 7189.55 41.21 C Low 1.17 One 0.5 miles upstream Aggressive 296700.40 7159.16 41.44 D Low 1.17 One 1.0 mile upstream Moderate 297058.70 7211.75 41.19 E Low 1.17 Two 1.0 mile upstream Moderate 296866.60 7330.05 40.50 14 A Medium 1.17 One 0.5 miles upstream Moderate 297940.40 7378.97 40.38 B Medium 0.32 One 0.5 miles upstream Moderate 298161.10 7436.17 40.10 C Medium 1.17 One 0.5 miles upstream Aggressive 297940.40 7378.97 40.38 D Medium 1.17 One 1.0 mile upstream Moderate 297626.00 7452.88 39.93 E Medium 1.17 Two 1.0 mile upstream Moderate 298123.40 7483.30 39.84 15 A High 1.17 One 0.5 miles upstream Moderate 298536.60 7581.99 39.38 B High 0.32 One 0.5 miles upstream Moderate 298210.10 7644.96 39.01 C High 1.17 One 0.5 miles upstream Aggressive 298536.60 7581.99 39.38 D High 1.17 One 1.0 mile upstream Moderate 298593.80 7546.08 39.57 E High 1.17 Two 1.0 mile upstream Moderate 298713.40 7649.36 39.05

PAGE 127

127 Table 5 6. Cumulative network statistics on the I 95 in Miami during the concurrent operation of the Fuzzy Logic ramp metering and the volume based VSL Scenario # Demand (veh/hr) FUZZY Rule Weights Number of VSL signs Locations of VSL signs VSL operation Travel Distance Total (miles) Total Travel Time (Hours) Average Network Speed (mph) 16 A Low Proposed One 0.5 miles upstream Moderate 296448.90 7206.01 41.14 B Low Modified One 0.5 miles upstream Moderate 296792.00 7216.05 41.13 C Low Proposed One 0.5 miles upstream Aggressive 296448.90 7206.01 41.14 D Low Proposed One 1.0 mile upstream Moderate 296565.70 7038.61 42.13 E Low Proposed Two 1.0 mile upstream Moderate 296918.30 7079.09 41.94 17 A Medium Proposed One 0.5 miles upstream Moderate 297614.30 7367.87 40.39 B Medium Modified One 0.5 miles upstream Moderate 298562.00 7848.74 38.04 C Medium Proposed One 0.5 miles upstream Aggressive 297614.30 7367.87 40.39 D Medium Proposed One 1.0 mile upstream Moderate 297987.00 7446.92 40.02 E Medium Proposed Two 1.0 mile upstream Moderate 297615.50 7361.46 40.43 18 A High Proposed One 0.5 miles upstream Moderate 297414.00 7424.64 40.06 B High Modified One 0.5 miles upstream Moderate 297903.70 7675.66 38.81 C High Proposed One 0.5 miles upstream Aggressive 297414.00 7424.64 40.06 D High Proposed One 1.0 mile upstream Moderate 297829.30 7430.56 40.08 E High Proposed Two 1.0 mile upstream Moderate 297504.50 7315.37 40.67

PAGE 128

128 Table 5 7. Cumulative network statistics on the I 95 in Miami during the concurrent operation of the Fuzzy Logic ramp metering and the occupancy based VSL Scenario # Demand (veh/hr) FUZZY Rule Weights Number of VSL signs Locations of VSL signs VSL operation Travel Distance Total (miles) Total Travel Time (Hours) Average Network Speed (mph) 19 A Low Proposed One 0.5 miles upstream Moderate 296686.40 7048.60 42.10 B Low Modified One 0.5 miles upstream Moderate 296313.50 7187.54 41.23 C Low Proposed One 0.5 miles upstream Aggressive 296686.40 7048.60 42.10 D Low Proposed One 1.0 mile upstream Moderate 296432.00 6946.19 42.68 E Low Proposed Two 1.0 mile upstream Moderate 296643.30 6999.68 42.38 20 A Medium Proposed One 0.5 miles upstream Moderate 297552.00 7234.36 41.13 B Medium Modified One 0.5 miles upstream Moderate 297605.50 7524.46 39.55 C Medium Proposed One 0.5 miles upstream Aggressive 297552.00 7234.36 41.13 D Medium Proposed One 1.0 mile upstream Moderate 297583.80 7396.32 40.23 E Medium Proposed Two 1.0 mile upstream Moderate 297683.10 7292.97 40.82 21 A High Proposed One 0.5 miles upstream Moderate 297072.50 7385.51 40.22 B High Modified One 0.5 miles upstream Moderate 298725.80 7473.98 39.97 C High Proposed One 0.5 miles upstream Aggressive 297072.50 7385.51 40.22 D High Proposed One 1.0 mile upstream Moderate 297289.80 7635.93 38.93 E High Proposed Two 1.0 mile upstream Moderate 297999.50 7276.28 40.95

PAGE 129

129 Table 5 8. Results of the conducted statistical analysis. Compared Control Cases Difference of average network speed values ( 1 2 ) Standard Deviation (s d ) value VSL Control No Control (33.17 31.98) = 1.18 1.31 2.87 1.833 Ramp Metering Control No Control (39 83 31.98) = 7.85 2.40 10.34 1.833 Integrated Control No Control ( 40 95 31.98) = 8.97 2.44 11.65 1.833 Ramp Metering Control VSL Control (3 9 83 3 3 17 ) = 6.66 1.94 10.84 1.833 Integrated Control VSL Control ( 40 95 3 3 17 ) = 7.78 1.90 12.93 1.833 Integrated Control Ramp Metering Control ( 40.95 3 9 83 ) = 1.1 2 1.86 1.91 1.833

PAGE 130

130 Figure 5 1. Average speed per link under four different traffic control schemes for the low demand scenario between 16:15 16:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 131

131 Figure 5 2. Average speed per link under four different traffic control schemes for the low demand scenario between 16:45 17:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 132

132 Figure 5 3. Average speed per link under four different traffic control schemes for the low demand scenario between 17:15 17:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 133

133 Figure 5 4. Average speed per link under four different traffic control schemes for the low demand scenario between 17:45 18:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 134

134 Figure 5 5. Average speed per link under four different traffic control schemes for the low demand scenario between 18:15 18:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 135

135 Figure 5 6. Throughput north of Opa Locka Blvd. under all the examined traffic management schemes for the low demand scenario. 1200 1300 1400 1500 1600 1700 1800 1900 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 136

136 Figu re 5 7. Throughput north of Biscayne Canal under all the examined traffic management schemes for the low demand scenario. 1200 1300 1400 1500 1600 1700 1800 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 137

137 Figure 5 8. Travel time per vehicle from I 195 EB/WB till the exit to 95 th St. under all the examined traffic management scheme s for the low demand scenario. 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 138

138 Figure 5 9. Travel time per vehicle from NW 81 st St. till the exit to North Golden Glades Dr. under all the examined traffic management schemes for the low demand scenario. 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 22.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL SL NO CONTROL

PAGE 139

139 Figure 5 10. Delay per vehicle on the on ramp from NW 125 th St. under all the examined traffic management schemes for the low demand scenario. 0.00 0.25 0.50 0.75 1.00 1.25 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & OCC VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 140

140 Figure 5 11. Delay per vehicle on the on ramp from Opa Locka Blvd. under all the examined traffic management schemes for the low demand scenario. 0.00 1.00 2.00 3.00 4.00 5.00 6.00 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 141

141 Figure 5 12. Average speed per link under four different traffic control schemes for the medium demand scenario between 16:15 16:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 142

142 Figure 5 13. Average speed per link under four different traffic control sche mes for the medium demand scenario between 16:45 17:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 143

143 Figure 5 14. Average speed per link under four different traffic control schemes for the medium demand scenario between 17:15 17:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 144

144 Figure 5 15. Average speed per link under fou r different traffic control schemes for the medium demand scenario between 17:45 18:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 145

145 Figure 5 16. Average speed per link under four different traffic control schemes for the medium demand scenario between 18:15 18:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 146

146 Figure 5 17. Throughput north of Opa Locka Blvd. under all the examined traffic management schemes for the medium demand scenario. 1200 1300 1400 1500 1600 1700 1800 1900 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 147

147 Figure 5 18. Throughput north of Biscayne Canal under all the examined traffic management schemes for the medium demand scenario. 1200 1300 1400 1500 1600 1700 1800 1900 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 148

148 Figure 5 19. Throughput south of US 441 for the best implementation of each traffic management scheme during the medium demand scenario. 800 900 1000 1100 1200 1300 1400 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) No Control VSL Only Ramp Metering Only Ramp Metering & VSL

PAGE 149

149 Figure 5 20 Travel time per vehicle from I 195 EB/WB till the exit to 95 th St. under all the examined traffi c management schemes for the medium demand scenario. 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 150

150 Figure 5 2 1 Travel time per vehicle from NW 81 st St. till the exit to North Golden Glades Dr. under all the examined traffic management schemes for the medium demand scenario. 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 22.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 151

151 Figure 5 2 2 Delay per vehicle on the on ramp from NW 125 th St. under all the examined traffic management schemes for the medium demand scenario. 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 152

152 Figure 5 2 3 Delay per vehicle on the on ramp from Opa Locka Blvd. under all the examined traffic management schemes for the medium demand scenario. 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 5.50 6.00 6.50 7.00 7.50 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & VOL VSL ALINEA & OCC VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 153

153 Figure 5 2 4 Average speed per link under four different traffic control schemes for the high demand scenario between 16:15 16:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 154

154 Figure 5 2 5 Average speed per link under four different traffic control sch emes for the high demand scenario between 16:45 17:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL OCC VSL

PAGE 155

155 Figure 5 2 6 Average speed per link under four different traffic control schemes for the high demand scenario between 17:15 17:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 156

156 Figure 5 2 7 Average speed per link under four d ifferent traffic control schemes for the high demand scenario between 17:45 18:00 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 157

157 Figure 5 2 8 Average speed per link under four different traffic control schemes for the high demand scenario between 18:15 18:30 pm. 0 10 20 30 40 50 60 70 Speed (mph) ALINEA & OCC VSL FUZZY & OCC VSL OCC VSL NO CONTROL

PAGE 158

158 Figure 5 2 9 Throughput north of Opa Locka Blvd. under all the examined traffic management schemes for the high demand scenario. 1200 1300 1400 1500 1600 1700 1800 1900 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 159

159 Figure 5 30 Throughput north of Biscayne Canal under all the examined traffic management schemes for the high demand scenario. 1200 1300 1400 1500 1600 1700 1800 1900 1 2 3 4 5 6 7 8 9 10 11 12 Throughput (vphpl) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 160

160 Figure 5 3 1 Travel time per vehicle from I 195 EB/WB till the exit to 95 th St. under all the examined traffic management schemes for the high demand scenario. 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 16.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 161

161 Figure 5 3 2 Travel time per vehicle from NW 81 st St. till the exit to North Golden Glade s Dr. under all the examined traffic management schemes for the high demand scenario. 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 20.00 22.00 24.00 26.00 28.00 1 2 3 4 5 6 7 8 9 10 11 12 Travel Time (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 162

162 Figure 5 3 3 Delay per vehicle on the on ramp from NW 125 th St. under all the examined traffic management schemes for the high demand scenario. 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 163

163 Figure 5 3 4 Delay per vehicle on the on ramp from Opa Locka Blvd. under all the examined traffic management schemes for the high demand scenario. 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 5.50 6.00 6.50 7.00 7.50 1 2 3 4 5 6 7 8 9 10 11 12 Delay per Vehicle (min) Time Period (15 min) ALINEA & OCC VSL ALINEA & VOL VSL FUZZY & OCC VSL FUZZY & VOL VSL ALINEA FUZZY OCC VSL VOL VSL NO CONTROL

PAGE 164

164 Figure 5 3 5 One VSL sign located just downstream of the entrance from Opa Locka Boulevard.

PAGE 165

165 Figure 5 36 One VSL sign located immediately upstream of the entrance from Opa Locka Boulevard.

PAGE 166

166 Figure 5 37 Installation of two VSL signs on the northbound direction of the I 95 in Miami.

PAGE 167

167 Figure 5 3 8 Interactions between ramp metering and VSL during Scenario 16 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 168

168 Figure 5 3 9 Interactions between ramp metering and VSL during Scenario 16 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 169

169 Figure 5 40 Interactions between ramp metering and VSL during Scenario 16 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 170

170 Figure 5 4 1 Interactions between ramp metering and VSL during Scenario 16 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 171

171 Figure 5 4 2 Interactions between ramp metering and VSL during Scenario 19 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 172

172 Figure 5 4 3 Interactions between ramp metering and VSL during Scenario 19 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (min) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 173

173 Figure 5 4 4 Inte ractions between ramp metering and VSL during Scenario 19 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 174

174 Figure 5 4 5 Interactions between ramp metering and VSL during Scenario 19 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 175

175 Figure 5 4 6 Interactions between ramp metering and VSL during Scenario 10 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 176

176 Figure 5 47 Interact ions between ramp metering and VSL during Scenario 10 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 177

177 Figure 5 4 8 Interactions between ramp metering and VSL during Scenario 10 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 178

178 Figure 5 4 9 Interactions between ramp metering and VSL during Scenario 10 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 60 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 179

179 Figure 5 50 Interactions between ramp metering and VSL during Scenario 13 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 180

180 Figure 5 5 1 Interactions between ramp metering and VSL during Scenario 13 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 181

181 Figure 5 5 2 Interactions between ramp metering and VSL during Scenario 13 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 182

182 Figure 5 5 3 Int eractions between ramp metering and VSL during Scenario 13 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 183

183 Figure 5 5 4 Interactions between ramp metering and VSL during Scenario 17 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (Veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 184

184 Figure 5 5 5 Interactions between ramp metering and VSL during Scenario 17 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 185

185 Figure 5 5 6 Interactions between ramp metering and VSL during Scenario 17 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 186

186 Figure 5 5 7 Interactions between ramp metering and VSL during Scenario 17 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min)) Simulation Time line (min) Metering Rate VSL Rate

PAGE 187

187 Figure 5 5 8 Interactions between ramp metering and VSL during Scenario 20 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 188

188 Figure 5 5 9 Interactions between ramp metering and VSL during Scenario 20 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 189

189 Figure 5 60 Interactions between ramp metering and VSL during Scenario 20 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 190

190 Figure 5 6 1 Interactions between ramp metering and VSL during Scenario 20 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 191

191 Figure 5 6 2 Interactions between ramp metering and VSL during Scenario 11 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (Veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 192

192 Figure 5 63 Interactions between ramp metering and VSL during Scenario 11 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 193

193 Figure 5 6 4 Interactions between ramp metering and VSL during Scenario 11 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (Veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 194

194 Figure 5 6 5 Int eractions between ramp metering and VSL during Scenario 11 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) ALINEA & OCC VSL ALINEA & OCC VSL

PAGE 195

195 Figure 5 6 6 Interactions between ramp metering and VSL during Scenario 14 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 196

196 Figure 6 6 7 Interactions between ramp metering and VSL during Scenario 14 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 197

197 Figure 5 6 8 Interac tions between ramp metering and VSL during Scenario 14 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 198

198 Figure 5 6 9 Interactions between ramp metering and VSL during Scenario 14 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 199

199 Figure 5 70 Interactions between ramp metering and VSL during Scenario 18 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 200

200 Figure 5 7 1 Interactions between ramp metering and VSL during Scenario 18 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 201

201 Figure 5 7 2 Interactions between ramp metering and VSL during Scenario 18 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 202

202 Figure 5 7 3 Interactions between ramp metering and VSL during Scenario 18 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 203

203 Figure 5 7 4 Int eractions between ramp metering and VSL during Scenario 21 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 204

204 Figure 5 7 5 Interactions between ramp metering and VSL during Scenario 21 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 205

205 Figure 5 7 6 Interactions between ramp metering and VSL during Scenario 21 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 206

206 Figure 5 7 7 Interactions between ramp metering and VSL during Scenario 21 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 207

207 Figure 5 7 8 Interactions between ramp metering and VSL during Scenario 12 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 208

208 Figure 5 7 9 Interactions between ramp metering and VSL during Scenario 12 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 209

209 Figure 5 80 Interactions between ramp metering and VSL during Scenario 12 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 210

210 Figure 5 8 1 Interactions between ramp metering and VSL during Scenario 12 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 211

211 Figure 5 8 2 Interactions between ramp metering and VSL during Scenario 15 A. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 212

212 Figure 5 83 Interactions between ramp metering and VSL during Scenario 15 B. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 213

213 Figure 5 8 4 Interactions between ramp metering and VSL during Scenario 15 D. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 214

214 Figure 5 8 5 Interactions between ramp metering and VSL during Scenario 15 E. 20 25 30 35 40 45 50 55 0 5 10 15 20 25 30 35 40 45 50 55 1 15 30 45 60 75 90 105 120 135 150 165 180 VSL Rate (mph) Metering Rate (veh/min) Simulation Time line (min) Metering Rate VSL Rate

PAGE 215

215 CHAPTER 6 SUMMARY AND CONCLUSI ONS Ramp metering is a freeway traffic management strategy which was developed and widely implemented over the past fifty years. Since the debut of ramp metering in the early 1960s, several metering algorithms have been developed. The mode of operation, level of sophistication and performance objectives of the existing algorithms varies significantly. Ramp metering has been thoroughly evaluated through field implementations and simula tion analyses. All the relevant studies demonstrate that the deployment of ramp metering results in an efficient exploitation of the freeway capacity. It has been shown that ramp metering not only postpones the activation of bottlenecks, but even manages t o sustain higher average discharge queue flow rate after the activation compared to the unmetered case (Zhang and Levinson, 2010). VSL is another notable freeway traffic management scheme which has been also widely used in the U.S.A. and abroad. VSL have b een mostly installed around work zones and freeway locations where recurring congestion occurs. It has been shown that VSL diminish shockwaves upstream of bottlenecks by stabilizing traffic flow (i.e. reducing speed differences among vehicles) (Smulders, 1 990). However, simulation analyses and evaluations of field implementations of VSL have been inconclusive so far regarding the effects of the strategy on aggregate traffic flow. It is not clear whether VSL can increase average speed and throughput along th e freeway. The assessment of the combined operation of ramp metering and VSL has been no studies that have examined the simultaneous implementation of both strategies i n a freeway facility yet. A macroscopic deterministic simulation tool named METANET has

PAGE 216

216 been mostly utilized for the analysis of the concurrent operation of ramp metering and VSL. These studies have indicated that the performance of the integrated control is far superior to other types of control (i.e. ramp metering only, VSL only) with respect to network wide statistics (Heg yi et al., 2004, Papamichail et al., 2008, Ghods et al. 2009, Carlson et al., 2010). However, the networks considered in these analyse s are hypothetical and the demand scenarios relatively simple. This work evaluates the concurrent operation of ramp metering and VSL with the use of CORSIM; a stochastic microscopic simulation tool. The network considered is the northbound direction of the I 95 in Miami. Traffic data collected from the STEWARD database are used for the replication of the actual traffic conditions on the simulated network through calibration. The performance of the integrated control scheme is compared against the no control the VSL only and the ramp metering only cases. A comparison is made with respect to network wide and link statistics. Traffic conditions are evaluated both on the mainline network and the on ramps. Moreover, this analysis examined the interactions of the two systems when implemented together. The ramp metering algorithms considered are the ALINEA ramp metering algorithm and the Fuzzy Logic ramp metering algorithm. These algorithms were combined with two VSL algorithms; an occupancy based VSL algorithm and a volume based VSL algorithm. The four different combinations were evaluated at three different demand levels; low, medium and high. Each combination was simulated for different calibration parameters of the metering algorithm, for different VSL sign posi tioning and different thresholds determining the operation of the VSL.

PAGE 217

217 However, the Fuzzy Logic ramp metering algorithm and the two VSL algorithms s were built using C++ code in order to replicate the operation of these algorithms and eventually interface them with CORSIM. The RTE replicating the Fuzzy Logic ramp metering algorithm was developed exclusively for this study. On the contrary, the two VSL algorithms had been previously developed by Letter ( 2011) for the individual examination of VSL. Although the ALINEA ramp metering algorithm is integrated into CORSIM, the tool is incapable of providing the metering rates as an output. Since the metering rates were necessary for the evaluation of the intera ctions of the two systems, the ALINEA algorithm was also built as an RTE that included the functionality to provide the metering rates as an output. The following were concluded: During the no control case traffic conditions on the mainline network deterio rated significantly as demand increased from low to high. However, traffic operations on the ramps were not severely influenced. The performance of the VSL is dependent on the type of algorithm applied, number and location of the VSL signs, but not on the different thresholds determining the operation of the VSL. If demand increases on the network VSL should be operational along a longer stretch of the freeway to improve operations. No specific VSL algorithm was found to perform better at all three demand levels. During the medium demand scenario the volume based VSL outperformed the occupancy based VSL. However, at the other two demand levels it was the occupancy based VSL algorithm that exhibited a better performance. The benefits from implementing only V SL along the freeway are mostly realized during the medium demand scenario. Traffic conditions improve marginally compared to the no control case when demand is low or high. Ramp metering regulates traffic very effectively along the mainline network even d uring high demand conditions. Travel times in the network slightly increase when demand increases from low to high. However, the improvement of traffic conditions on the mainline occurs at the expense of spill back on the ramps. As demand increases queues on the on ramps also increase.

PAGE 218

218 The Fuzzy Logic ramp metering algorithm outperforms the ALINEA algorithm irrespective of the demand level. The main difference between the two algorithms is that the Fuzzy Logic algorithm is balancing traffic between the main line and the on ramps more effectively. Thus, both queues on the ramp and on the mainline are shorter. The ramp metering only control sustains a substantially better performance on the freeway compared to the VSL only case. This should be partially ascribe d to the fact that a small portion of the network is VSL operated. On the other hand, most of the existing on ramps along the freeway are metered. The most significant improvement to the traffic conditions on the I 95 in Miami occurs when the integrated co ntrol scheme is deployed. Travel time along the mainline decreases significantly relative to the other control cases. Queues are managed more efficiently compared to the metering only case. The improvement is also depicted in the network wide statistics. T he highest possible average network speed is attained when the Fuzzy Logic ramp metering algorithm and the occupancy based VSL operate concurrently. The combination of Fuzzy Logic ramp metering and VSL outperforms that of ALINEA metering and VSL. This occu rs irrespective of the demand scenario. When the Fuzzy Logic ramp metering is combined with VSL queues fully dissipate on the freeway by the end of the simulation. This is the only scenario that this phenomenon is observed. Throughput increases significant ly relative to the no control case when any of the control strategies is applied. It might even increase up to 50veh/15min upstream of the bottleneck location. However, it is clear that there are minor differences with respect to throughput between the VSL only, the ramp metering only and the integrated control case. During the integrated control scheme it is always the VSL system that begins to restrict traffic first. This occurs because congestion initially breaks out around the location of the detectors that determine the operation of the VSL. The response of ramp metering to the restrictive operation of the VSL depends on the location of the VSL signs. The furthest upstream the VSL sign is located from the on ramp the more time is required for the meter ing algorithm to become restrictive. The response of the VSL to the changes in the operation of the ALINEA ramp metering algorithm depends on the value of the ALINEA calibration parameter. The higher the value of the calibration parameter the earlier the response of the VSL will be. During the operation of the volume based VSL the VSL rate never attains its lowest possible value (i.e. 40mph). Moreover, the operation of the VSL in this case is rather oscillatory and unstable.

PAGE 219

219 The interface of the RTEs wit h CORSIM is a rather tedious, time consuming and prone to mistakes task. Thus, it is recommended that the unavailable algorithms are embedded into CORSIM and a new interface is developed to allow the manipulation of the settings of these algorithms. Recomm endations regarding a new interface include the following: Provide an input screen for each type of algorithm. Through this input screen allow the user to modify the fuzzy rule weights for the Fuzzy Logic ramp metering algorithm and the thresholds determin ing the speed limit changes for the VSL algorithms. Allow the user to place meters or VSL signs along the network as objects. Provide a visual interface to facilitate the connection of VSL or meters with specific detectors. Develop a new output screen that will provide the option of outputting the metering and the VSL rates. Recommendations regarding future research include the following: Currently the updated speed limits (i.e. VSL case) are enforced on a link basis. However, in practice not every vehicle complies with the VSL. Thus, it would be preferable for the speed limit changes to occur on a vehicle basis instead. The VSL operated portions of the network should be extended to consider additional bottlenecks along the I 95 in Miami. It is recommended t hat the integrated control scheme is implemented in the field, so that a comparison of simulation and actual results contributes to the validation and improvement of the simulation tool. The interactions between ramp metering and VSL should be evaluated in conjunction with demand and congestion indicators. O ccupancy should be measured at the detectors that dictate the changes of the speed limits.

PAGE 220

220 APPENDIX A CALIBRATION DATA This appendix provides details regarding the inputs and c alibration of the CORSIM file. Table A 1 includes information regarding the entering traffic in the simulated network along with the exiting percentag es at each off ramp during every time interval. Table A 2 shows the values of the car following sensitivity factor and f ree flow speed (mph) for each network link.

PAGE 221

221 Table A 1. Entering and exiting percentages in the simulated ne twork.

PAGE 222

222 Table A 2 Car following sensitivity parameter by link Link # Free Flow Speed (mph) Car following Sensitivity Multiplier Link # Free Flow Speed (mph) Car following Sensitivity Multiplier [105, 106] 60 1 .00 [139, 140] 60 1.65 [106, 107] 60 1 .00 [140, 141] 55 1.5 0 [107, 108] 55 1.00 [141, 142] 55 1.00 [108, 109] 55 1.00 [142, 145] 50 1.50 [109, 111] 55 1.00 [145, 146] 50 1.80 [111, 112] 55 2.00 [146, 147] 50 1.00 [112, 113] 55 2.00 [147, 148] 50 1.00 [113, 114] 60 1.00 [148, 149] 50 1.00 [114, 115] 60 1.00 [149, 150] 50 1.00 [115, 116] 60 1.00 [150, 152] 50 1.00 [116, 117] 60 1.00 [152, 153] 50 1.50 [117, 118] 60 1.00 [153, 154] 50 1.65 [118, 119] 60 1.00 [154, 155] 50 1.50 [119, 120] 60 1.00 [155, 156] 50 1.50 [120, 121] 60 1.00 [156, 157] 50 1.50 [121, 122] 60 1.00 [157, 159] 50 1.80 [122, 123] 60 1.00 [158, 159] 50 1.80 [123, 124] 60 1.00 [159, 161] 55 1.80 [124, 125] 60 1.00 [161, 165] 55 1.80 [125, 126] 60 1.00 [165, 166] 60 1.00 [126, 127] 60 1.00 [166, 167] 60 1.00 [127, 128] 60 1.00 [167, 168] 60 1.00 [128, 129] 60 1.60 [168, 170] 60 1.00 [129, 130] 60 1.60 [170, 171] 60 1.00 [130, 131] 60 1.70 [171, 172] 60 1.00 [131, 132] 60 1.90 [172, 174] 60 1.00 [132, 133] 60 2.00 [174, 177] 60 1.20 [133, 134] 60 1.50 [177, 178] 60 1.20 [134, 135] 60 1.00 [178, 179] 60 1.00 [135, 136] 60 1.00 [179, 180] 60 1.00 [136, 137] 60 1.00 [180, 181] 60 1.20 [137, 138] 60 1.00 [181, 182] 60 1.00 [138, 139] 60 1.65 [182, 185] 60 1.00

PAGE 223

223 LIST OF REFERENCES Allaby P., Hellinga B., Bullock M., 2007. Variable Speed Limits: Safety and Operational Impacts of a Candidate Control Strategy for Freeway Applications. IEEE Transactions on In telligent Transportation System 8 (4), 671 680. Carlson, R., Papami chail, I., Papageorgiou, M., Messmer, A., 2 010. Optimal Motorway Traffic Flow Control Involving Variable Speed Limits and Ramp Metering. Transportati on Science 44 ( 2 ) 238 253. Cremer, M., 1979. Der Verkehrsflu Springer, Berlin. El efteriadou, L., Roess, R.P., McShane, W.R., 1 995. The Probabilistic Nature of Breakdown at Freeway Merge Junctions. Transportation Research Record 1484, National Academy P ress, 80 89. Elefteriadou, L., Kondyli, A., Brilon, W., Washburn S S ., Hall, F., Persaud, B., 2009. Proactive Ramp Management U nder the Threat of Freeway Flow Breakdown. Final Report NCHRP 3 87. Freeway Management and Operations Handbook, 2003. FHWA OP 04 003. Ghods, A., Kian, A., Tabibi, M., 2009. Adaptive Freeway Ramp Metering and Variable Speed Limit Control: A Genetic Fuzzy A pproach. IEEE Intelligent Transportation Syste ms Magazine 1 ( 1 ) 27 36. Haas R., Carter M., Perry E., Trombly J., Bedsole E., and Margiotta R., 2009. iFlorida Model Deployment Final Evaluation Report, Washington, D.C. FHWA HOP 08 050. Hegyi, A., 2004. Model predictive control for integrating traffic control measures. Ph.D. thesis, TRAIL Thesis Series T2004/2, Delft University of Technology, Delft The Netherlands. Hegyi, A., De Schutter, B., Hellendoorn, H., 2005. Model predictive control for optimal coordination of ramp meterin g and variable speed limits. Transportation Research Part C 13 ( 3 ) 185 209. Highway Capacity Manual (HCM), 2000. Transportation Research Board. Washington, DC, ISBN: 0 309 06681 6. Hines, M., 2002. Judicial Enforcement of Variable Speed Limits. NCHRP Legal Research Digest, 47. Kwon, E., Brannan, D., Shouman, K., Isackson, C., Arseneau, B., 2007. Development and Field Evaluation of Variable Advisory Speed Limit System for Work Zones. Transportation Research Re cord 2015, 12 18.

PAGE 224

224 Lee, C., Hellinga, B., Saccomanno, F., 2004. Assessing Safety Benefit s of Variable Speed Limits. Trans portation Research Record 1897, 183 190. Letter, C., 2011. A Framework for Simulating Variable Speed Limit Algorithms in CORSIM. Master Thesis, University of Florida, Gainesville, USA. Meldrum, D. Taylor, C., 2000. Algorithm Design, User Interface, and Optimization Procedure for a Fuzzy Logic Ramp Metering Algorithm: A Training Manual for Freeway Operations Engineers Technical Report to Washington State Department of Transportation, Seattle. Messmer, A., Papageorgiou, M., 1990. METANET: a macroscopic simulation program for motorway networks. Traffic Engineering Control 31, 466 470. Papageorgiou, M., Hadj Salem, H., Middelham, F., 1997. ALINEA Local Ramp Metering Summary of Field Results. Trans portation Research Record 1603, 90 98. Papageorgiou, M., Kosmatopoulos, E., Papamichail, I., 2008. Effects of variable speed li mits on motorway traffic flow. Transpor tation Research Record 2047, 3 7 48. Papageorgiou M., Papamichail, I., Messmer, A., Wang, Y., 2010. Traff ic Simulation with METANET. Fundamentals of Traffic Simulation, International Series in Operations R esearch and Management Science 145, 399 430. Papamichail, I., Kamp itaki, K., Papa georgiou, M., Messmer, A., 2008. Integrated Ramp Metering and Variable Speed Limit Control of Motorway Traffic Flow. In: Proceedings of the 17 th World Congress, The International Federation of Automatic Control, Seoul, Korea. PBS&J., 2009. I 4 Variable Spe ed Limit Effectiveness Study. Prepared for the Florida Department of Transportation, District 5. Ramp Management and Control Handbook, 2006. FHWA HOP 06 001. Robinson, M. D., 2000. Examples of Variable Speed Application. In: Proceedings of the 79th Annual Meeting of the Transportation Research Board, Washington, D.C. Sisiopiku, V., 2001. Variable Speed Control: Technology and Practices. In: Proceedings of the 11 th Annual Meeting of ITS America. Smaragdis, E., Papageorgiou, M., 2003. Series of New Loca l Ramp Metering Strategies. Trans portation Research Record 1856, 74 86.

PAGE 225

225 Smulders, S., 1990. Control of freeway traffic f low by variable speed signs. Transportation Research Part B 24 (2), 111 132. Zackor H., 1991. Speed limitation on freeways: Traffic responsi ve strategies. In: Concise Encyclopedia of Traffic and Transpor tation Systems, Oxford, UK, 507 511. Zhang, M., Kim, T., Nie, X., Jin, W., Chu, L., Recker, W., 2001. Evaluation of On Ramp Control Algorithms. California PATH Research Report UCB ITS PRR 2001 36. Zhang, L., Levinson, D., 2010. Ramp metering and freeway bottleneck c apacity. Transportation Research P art A: Policy and Practice, 218 235.

PAGE 226

226 BIOGRAPHICAL SKETCH Evangelos Mintsis was born in Edessa, Greece in 1984. In 2009 h e received his Bachelor of Science in civil engineering from the Aristotle University of Thessaloniki, Greece. In 2012, he earned his Master of Science in civil engineering from the University of Florida. During his graduate studies, he was a research assi stant for his advisor, Dr. Lily Elefteriadou.