A Framework for Simulating Variable Speed Limit Algorithms in Corsim

MISSING IMAGE

Material Information

Title:
A Framework for Simulating Variable Speed Limit Algorithms in Corsim
Physical Description:
1 online resource (185 p.)
Language:
english
Creator:
Letter,Clark
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 Co-Chair:
Washburn, Scott S
Committee Members:
Yin, Yafeng

Subjects

Subjects / Keywords:
variable -- 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:
A major problem associated with freeway operations around major cities is congestion occurrence during peak volume periods. Typically, bottlenecks at merging and diverging junctions as well as incidents create a shockwave that propagates upstream. One of the tools currently examined as a way to dampen the shockwave produced by this bottleneck is variable speed limits (VSL). Current micro-simulators do not provide an interface to easily simulate VSLs and evaluate their impact on traffic, thus simulation must be carried out through additional coding. This study creates a test-bed for simulating and evaluating multiple VSL algorithms using the Corridor Simulation (CORSIM) micro-simulator. Three algorithms for VSL control are selected and simulated to evaluate the effectiveness of each algorithm. The roadway used for the simulation is a 13-mile section of I-95 in Miami, Florida. A run-time extension (RTE) interface is built to communicate with the CORSIM simulation and replicate the VSL operations. Different threshold values are tested to evaluate the effectiveness of each algorithm under various settings. It was concluded that all but one of the scenarios tested show an improvement in the average travel speed and total travel time after VSL is implemented. The throughput for most scenarios showed an improvement when observed over the time duration of the congestion. Overall, the volume-based algorithm showed the most improvement in the simulations.
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 Clark Letter.
Thesis:
Thesis (M.S.)--University of Florida, 2011.
Local:
Adviser: Elefteriadou, Ageliki L.
Local:
Co-adviser: Washburn, Scott S.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2011
System ID:
UFE0043389:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 A FRAMEWORK FOR SIMULATING VARIABLE SPEED LIMIT ALGORITHMS IN CORSIM By CLARK LETTER A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DE GREE OF MASTE R OF SCIENCE UNIVERSITY OF FLORIDA 2011

PAGE 2

2 2011 Clark Letter

PAGE 3

3 To my family and friends

PAGE 4

4 ACKNOWLEDGMENTS I thank my advisor, Dr. Lily Elefteriadou, Professor, Department of Civil and Coastal Engineering, University of Florida, for her guidan ce 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, Assistant Professor, for their guidance and feedback on the study. I would like to give a spe cial thanks to Mr. Tom Simmerman, for his assistance in computer programming and understanding of internal Corridor Simulation ( CORSIM ) structures. Finally, I would like to thank my family and friends for their constant moral support and encouragement.

PAGE 5

5 T ABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 12 1.1 Background ................................ ................................ ................................ ....................... 12 1.2 Objectives ................................ ................................ ................................ ......................... 13 1.3 Thesis Outline ................................ ................................ ................................ ................... 13 2 LITERATURE REVIEW ................................ ................................ ................................ ....... 14 2.1 Implementation of variable speed limits ( VSL ) ................................ ............................... 14 2.1.1 Implementation in the USA ................................ ................................ .................... 14 2.1.2 Implement ation in Europe ................................ ................................ ...................... 19 2.2 VSL Simulation ................................ ................................ ................................ ................ 24 2.3 VSL Algorithms ................................ ................................ ................................ ................ 32 2.3. 1 Congestion and Safety Related Algorithms ................................ ........................... 32 2.3.2 Weather Related and Other Algorithms ................................ ................................ 34 2.4 Corridor Simulation ( CORSIM ) Simul ation Software ................................ ..................... 35 2.5 Literature Review Summary ................................ ................................ ............................. 36 3 METHODOLOGY ................................ ................................ ................................ ................. 39 3.1 Study Site Selection and Data Assembly ................................ ................................ .......... 39 3.2 Development of run time extensions ( RTE ) ................................ ................................ ..... 39 3.3 Testing and Analysis of Algorit hms in CORSIM ................................ ............................. 40 4 STUDY SITE DESCRIPTION ................................ ................................ ............................... 41 4.1 Study Site Selection ................................ ................................ ................................ .......... 41 4.2 Calibrated CORSIM Simulation ................................ ................................ ....................... 42 5 DEVELOPMENT OF THE RTE ................................ ................................ ........................... 47 5.1 Selection of Algorithms ................................ ................................ ................................ .... 47 5.1.1 Algorithm Based on Occupancy ................................ ................................ ............. 48 5.1.2 Algorithm Based on Flow ................................ ................................ ...................... 48

PAGE 6

6 5.1.3 Algorithm Based on a Logic Tree including Flow, Occupancy, and Average Speed ................................ ................................ ................................ ............................ 49 5.2 Sign Location Variations ................................ ................................ ................................ .. 49 5.3 Building of dynamic link librarie s (DLL s ) to communicate with the RTE interface ....... 50 6 I MPLEMENTATION AND ANALYSIS OF ALGORITHMS IN CORSIM ....................... 61 6.1 Implementation of R TE Scenarios in CORSIM ................................ ............................... 61 6.2 Analysis of the No Control Scenario ................................ ................................ ................ 62 6.3 Analysis of VSL Algorithm Performance ................................ ................................ ........ 62 6.3.1 Analysis of the Occupancy Based Algorithm ................................ ........................ 63 6.3.2 Analysis of the Volume Based Algorithm ................................ ............................. 64 6.3.3 Analysis of the Multiple Parameter Based Algorithm ................................ ........... 66 6.4 Summary of Analysis ................................ ................................ ................................ ....... 67 7 SUMMARY AND CONCLUSIONS ................................ ................................ ................... 111 APPENDIX A CALIBRATION RESULTS FOR I 95 NETWORK ................................ ........................... 114 B Sample Source Code for RTE ................................ ................................ ............................... 126 C Speed Profiles for scenarios tested ................................ ................................ ....................... 134 LIST OF REFERENCES ................................ ................................ ................................ ............. 182 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 185

PAGE 7

7 LIST OF TABLES Table page 2 1 Orlando, Florida I 4 variable speed limit ( VSL ) control thresholds ................................ .. 37 5 1 Occ upancy thresholds for displayed speed limits (scenario 1 ) ................................ .......... 52 5 2 Occupancy Threshol ds for Displayed Speed Limits (s cenario 2) ................................ ...... 52 5 3 Occupancy thresholds for displayed speed limits (scenario 3) ................................ .......... 52 5 4 Volume thresholds for displayed speed limits (scenario 1) ................................ ............... 53 5 5 Volume thresholds for displayed speed limits (scenario 2) ................................ ............... 53 6 1 Description of scenarios tested ................................ ................................ .......................... 70 6 2 Network performance measures for the occupancy based scenario ................................ .. 71 6 3. Occupancy thresholds for displayed speed limits (scenario 2) ................................ .......... 72 6 4 Network perf ormance measures for the volume based algorithm ................................ ..... 73 6 5 Network performance measures for the multiple parameter based algorithm ................... 7 3

PAGE 8

8 LIST OF FIGURES Figure page 2 1 Decision path for determining the new posted speed of the trigger variable speed limits ( VSLs ) ................................ ................................ ................................ ..................... 38 2 2 Operation of Corridor Simulation ( CORSIM ) and run time extensions ( RTEs ) ............... 38 4 1 Section of I 95 being simulated by CORSIM ................................ ................................ .... 43 4 2 Location of de tectors along the section of I 95 study site containing proposed VSL implementation ................................ ................................ ................................ .................. 44 4 3 Speed profile of the I 95 section for time period 4 ................................ ............................ 45 4 4 Screen shot of I 95 network represented in CORSIM ................................ ....................... 46 5 1 Decision tree logic for third algorithm ................................ ................................ ............... 54 5 2 Use of one sign spaced approximately one half mile ................................ ........................ 55 5 3 Use of one sign spaced approximately one mile. ................................ ............................... 55 5 4 Use of two signs spaced approximately one half mile apart ................................ ............. 56 5 5 Use of two signs spaced approximately one mile apart ................................ ..................... 56 5 6 Screen shot of text outputtin g to screen when speed changes ................................ ........... 57 5 7 Flowchart of general RTE logic ................................ ................................ ......................... 58 5 8 Time series of speed limit propagation down stream from VSL sign location .................. 59 6 1 Setting up the RTE tool configuration ................................ ................................ ............... 74 6 2 Specifying the call points of the RTE ................................ ................................ ................ 75 6 3 Speed profile for no control scenario over time the 12 periods. ................................ ........ 76 6 4 Speed profile for the occupancy based algorithm using two signs spaced mile apart (threshold scenario 2) compared to the no control scenario. ................................ .... 80 6 5 Throughput for the occupancy based algorithm using two signs spaced mile apart (threshold scenario 2) compared to the no control scenario. ................................ ............. 84 6 6 Speed profile for the volume based algorithm using one sign spaced 1 mile from the bottleneck source (threshold scenario 1) compared to the no control scenario. ................ 85

PAGE 9

9 6 7 Speed profile for the volume based algorithm using two signs spaced 1/2 mile apart (threshold scenario 1) compared to the no control scenario ................................ .............. 89 6 8 Throughput for the volume based algorithm using one sign spaced 1 mile from the bottleneck source (threshold scenario 1) compared to the no control scenario ................. 93 6 9 Th roughput for the volume based algorithm using two signs spaced 1/2 mile apart (threshold scenario 1) compared to the no control scenario. ................................ ............. 94 6 10 Logic tree thresholds used for the multiple par ameter algorithm ................................ ...... 95 6 11 Speed profile for the multiple parameter -based algorithm using two signs spaced 1/2 mile apart compared to the no control scenario. ................................ ................................ 96 6 12 Speed profile for the multiple parameter based algorithm using one sign spaced 1/2 mile from the bottleneck source compared to the no control scenario. ........................... 100 6 13 Throughput for the multiple parameter based algorithm using two signs spaced 1/2 mile apart compared to the no control scenario. ................................ .............................. 104 6 14 Throughput for the multiple parameter based algorith m using one sign spaced 1/2 mile from the bottleneck source compared to the no control scenario ............................ 105 6 15 Speed profile for the multiple parameter based algorithm using one sign spaced 1 mile fr om the bottleneck source compared to the no control scenario ............................ 106 6 16 Throughput for the multiple parameter based algorithm using one sign spaced 1 mile from the bottleneck source compared to t he no control scenario ................................ .... 110

PAGE 10

10 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 A FRAMEWORK FOR SIMU LATING VARIABLE SPEED LIMIT ALGORITHMS IN CORSIM By Clark Letter August 2011 Chair: Lily Elefteriadou Cochair: Scott Washburn Major: Civi l Engineering A major problem associated with freeway operations around major cities is congestion occurrence duri ng peak volume periods. Typically, b ottlenecks at merging and diverging junctions as well as incidents create a shockwave that propagates upstream. One of the tools currently examined as a way to dampen the shockwave produced by this bottleneck is variable speed limits (VSL). Current micro simulators do not provide an interface to easily simulate VSLs and evaluate their impact on traffic thus simulation must be carried out through additional coding. This study creates a test bed for simulating and evaluati ng multiple VSL algorithms using the Corridor Simulation ( CORSIM ) micro simulator. Three algorithms for VSL control are selected and simulated to evaluate the effectiveness of each algorithm. The roadw ay used for the simulation is a 13 m ile section of I 95 in Miami, Florida A run time extension (RTE) interface is built to communicate with the CORSIM simulation and replicate the VSL operations Di fferent threshold values are tested to evaluate the effectiveness of each algorithm under various settings It was concluded that all but one of the scenarios tested show an improvement in the average travel speed and total travel time after VSL is implemented. The throughput for

PAGE 11

11 most scenarios showed an improvement when observed over the time duration of the cong estion Overall, the volume based algorithm show ed the most improvement in the simulations

PAGE 12

12 CHAPTER 1 INTRODUCTION 1.1 Background Static speed limits are designed to provide motorists with a safe speed at which to drive. While these safe speeds are effe ctive during ideal conditions, they fail to provide recommended safe speeds during adverse weather or congested driving conditions (Sisiopiku 2001). Variable speed limits (VSLs) are a way of recommending safe driving speeds during less than ideal condition s. VSL systems have produced safety benefits such as a reduced number of rear end collisions and traffic homogenization. In addition to their safety benefits, VSL have been used upstream of bottlenecks with recurring congestion as a way to dampen the shoc kwave produced once congestion starts. However, the exact effects of VSL on traffic flow are not well understood, and the literature provides conflicting conclusions with respect to the effectiveness of VSL installations in increasing overall speeds and th roughput. W ith growing interest in identifying and implementing congestion mitigation techniques there is a need to simulate VSL and evaluate their potential impacts on traffic conditions in a more comprehensive manner Simulation is a very effective too l in evaluating alternatives under completely controlled conditions which cannot be achieved in the field. It is also very effective in providing a comprehensive picture of traffic operations in time and space. To date, f ew micro simulators possess the a bility to simulate VSL systems. Simulators such as AIMSUN and PARAMICS have the ability to simulate variable message signs, but require additional coding to simulate VSLs No micro simulator has a built in interface that allows the simulation of different VSL systems or algorithms. There are few tools or guidelines available for simulating VSLs, which makes simulating them a difficult and time consuming

PAGE 13

13 process Corridor Simulation ( CORSIM ) is a widely used micro simulator that does not have an interface to directly simulate VSLs. CORSIM has a run time extension (RTE) interface that allows users to define or modify operations of the simulation program. This interface can be used to simulate VSL operations and to test their effectiveness under a variety of co nditions and algorithm settings 1.2 Objectives The objective of this research is to develop a framework for simulat ing and testing VSL algorithm s in CORSIM through the use of an RTE Three different algorithms for VSL control are implemented into a CORSIM network and evaluated based on selected performance measures. The network analyzed is a 1 3 mile sect ion of I 95 in Miami, Florida This section was selected because the Florida Department of Transportation (FDOT) is interested in evaluating whether VSL wou ld be an effective strategy along this corridor. Different threshold values as well as several different VSL sign locations are tested with each algorithm to evaluate its effectiveness under different settings 1.3 Thesis Outline The remainder of this the sis is organized in 6 chapters. In Chapter 2, a literature review on the history of VSL operations both implemented and simulated VSL systems is presented. Chapter 3 discusses the methodology for carrying out the research. Chapter 4 presents the details of the selected study site and the CORSIM network. Chapter 5 explains the development of the RTE and the different algorithms tested. Chapter 6 presents the detailed procedures for performing the simulations, along with the analysis of results. Finally Chapt er 7 provides overall conclusions, and identifies direction s for future research.

PAGE 14

14 CHAPTER 2 LITERATURE REVIEW This literature review provides a summary of variable speed limit ( VSL ) implementation in the United States as well as in Europe It also descr ibes how VSL algorithms have been simulated using different software packages, enabling new and creative algorithms to be tested. Next, a summary of different algorithms is presented. Finally, an overview of the Corridor Simulation ( CORSIM ) software is pre sented, along with a discussion of the run time extension ( RTE ) function and its communication with CORSIM. 2.1 Implementation of VSL This section provides an overview of implemented VSL systems The section starts with a review of systems i mplemented in t he United States. The section then reviews implementations in Europe. 2.1.1 Implementation in the USA In the United States variable speed limits have been implemented in a number of locations. These systems typically set a safety speed limit according to t he weather, traffic, or road conditions ( McLawhorn, 2003). One of the most general uses of VSL s are at school zones and at construction or work zone s (Hines, 2002). The main objective of most freeway implementations in the US has been to improve safety, bu t few have focused on relief of traffic congestion. Congestion related benefits have been shown mostly using simulation. However, safety benefits have been documented for numerous VSL systems implemented in the field Abdel Aty et al., (2006) The first VS L system in the US was implemented along the M 10 (Lodge Freeway) in Detroit, Michigan, between the Edsel Ford Freeway (I 94) and the Davison Freeway in 1960. The system was designed to alert motorists to slow down when approaching congestion and

PAGE 15

15 accelerat e when leaving a congested area. The system was 3.2 miles long and had 21 VSL sign locations. The speed limits were chos en by the operator based on closed circuit television and plots of freeway speed. The VSL signs were manually switched at the control c e nter with an increment of 5 miles per hour (mph) from 20 to 60 mp h. The evaluation results show that the VSL system did not significantly increase or decrease vehicle speeds (Robinson, 2000). The system was disbanded sometime after 1967. In New Jersey, a V SL system was implemented along the New Jersey Turnpike in the 1960s. This system was designed to reduce speed limits during congested conditions, and is currently part of a larger ITS system. The system warns drivers of lane closures and crashes to improv e safety and avoid large delays. The system is ov er 148 miles in length and utilizes approximately 120 signs. Since the implementation of the system there have been updates to controllers and detectors to keep the system up to date and functioning effectiv ely The posted speed limits are based on average travel speeds and are displayed automatically. The posted speed limit can be reduced from the normal posted speed limit ( 65 mph 55 mph or 50 mph ) in increments of 5 mph to a minimum speed of 30 mph under six conditions: vehicle collisions, traffic congestion, construction, icy road conditions, snowfall, and fog. No formal evaluation of the system has been performed, but the Turnpike Authority observes the system 24 hours a day and has deemed its performanc e to be satisfactory. They did note that the system needed enforcement by State Police (Steel et al., 2005). In New Mexico, a VSL system was implemented along I 40 in Albuquerque in March of 1989 (Robinson 2002 ) The system was set up as a test bed for V SL equipment and was disbanded in 1997 due to road widening. The six kilometer long system used three roadside detector stations and a variable message sign to vary the posted speed limit. The posted speed

PAGE 16

16 limit was generated using a look up table based o n the smoothed (90 percent old data plus 10 percent current data) average speed plus a constant based on the environmental conditions. The speed and environmental data such as light level and precipi tation were collected by detectors. Evaluation results show that there was a slight reduction in accidents after the system was implemented. It has been suggested that the implementation of the National Maximum Speed Limit ( 55 mph ) hi ndered the effect of the system, as posted speeds were generated based on old match the expected conditions ( Steel et al., 2005). In Tennessee, a VSL system was implemented along a 19 mile section of I 75 in 1993 to respond to the reduction in visibility causing crashes during adverse weather con ditions (especially fog) (Robinson, 2002 ) The system has 10 VSL signs, 8 fog detector s 44 radar speed detectors, highway advisory radio and 6 swinging gates. The posted message and speed limit are determined by a central computer in the Highway Patrol o ffice, based on the transmitted data collected using environmental sensor and vehicle detectors. The system has the capability to close down the entire stretch of roadway during severe fog conditions, and divert traffic onto US Highway 11. This requires co ordination with highway patrol officers closing swinging gates. The effect of the VSL on actual travel speeds ha s not been formally evaluated, but the enforcement agency observed a slight (5 to 10 percent) reduction in speed, and there have been no crashes due to fog after the system was implemented. (Goodwin, 2003; Steel et al., 2005). In Colorado, a VSL system was implemented along the Eisenhower Tunnel on I 70 west of Denver in 1995. This system is designed to improve truck safety by displaying vehicle s pecific safe operating speeds for long downgrades. The system consists of a weigh in motion sensor, variable message sign, inductive loop detectors, and computer hardwar e and software. A safe speed is computed by an algorithm within the computer system bas ed on the truck weight, sp eed,

PAGE 17

17 and axle configuration. The recommended safe speed is then displayed on a variable message sig n. Moreover, each truck receives a vehicle specific recommended safe speed message. The speed limit was adviso ry and evaluation res ults show that truck related accidents declined on the steep downhill grade sections after the implementation of the VSL system, even though the truck volume increased (Robinson, 2000). The Washington Department of Transportation implemented a VSL system o n I 90 across the Snoqualmie Pass in 1997. The system was implemented to improve safety and inform motorists of road conditions and weather information and is still active. Speed limit s are recommended by the central computer based on information collected from a variety of sources, including wide aperture radar that tracks speeds, roadside cabinets that collect and control roadside data, and packetized data radio on three mountaintop relay sites that use microwave s to communicate to the control center. The computer automatically computes the speed from relayed data and recommends a VSL value, which an operator implements. It was found that VSLs may lose their effectiveness without enforceme nt by the State Patrol, and they reduced the mean speed and increase d the speed standard deviation ( Goodwin, 2003; Ulfarsson et al., 2001; Steel et al., 2005). In 1998, Northern Arizona University and the Arizona Departme nt of Transportation developed a VSL system based on a fuzzy control algorithm along the I 40 corridor in rural Arizona. This was an experimental system designed to display appropriate speeds for different weather conditions. It was unclear from the study whether the system was actually ever implemented, or just simulated. The system used a Road Weather I nformation System to gather atmospheric and road surface condi tions. The system then displayed a corresponding speed limit according to the fuzzy control algorith m. Placer (2001) summarized upgrades made to this Road

PAGE 18

18 Weather Information System. No perfor mance measures or quantitative impacts of the VSL system were given. In 2000 a VSL system was implemented along I 80 in Nevada. The system was remotely controlled without human intervention. It consisted of four VSL signs (two eastbound and two westbound ), visibility detectors, speed loops, R WIS weather stations, and d spe ed ahead signs upstream of the VSL signs. Speed limits were updated every 15 minutes and computed using a logic tree based on the 85th percentile speed, visibility, and pavement conditions. The results found that the sensors were unreliable and could not accurately relay visibility conditions (Robinson, 200 0; Robinson, 2002). This limited the effectiveness of the VSL system. No information was found on the current o perational status of the system. In Florida, VMS were placed along a 9 mile portion of I 4 in Orlando. The system is designed to improve safety along I 4 through more steady flow during congested periods, and to provide advance warning of slowing traffic ahead. Detectors are used to measure speed, volume, and occupancy for each lane at 30 second intervals. The SunGuide software monitor s the occupancy level and classifies traffic conditions as either free flow, light congestion, or heavy congestion. On the basis of these classifications, the software recommend s speed limits of 30 mph for heavy congestion, 40 mph for light congestion, and the normal speed limit (i.e., 50 or 55 mph) for free flow. The software also ensures that the posted speed limit does not change by more than 10 mph between two adjacent sets of VSL signs (Haas et al., 2009). A study prepared for the FDOT evaluated the performance of the current VSL operation (PBS&J, 2009). The study concluded that the VSL system was not effective at reducin g vehicle speeds. Since vehicles were not affected by the signs no traffic improvements or safety benefits were shown.

PAGE 19

19 A study was conducted i n southeast Wyoming (Buddemeyer, 20 10 ) to assess the effectiveness of VSL signs in a rural setting on a 100 mile stretch of I 80 through Elk Mountain. The system is designed to reduce speed limits during adverse weather conditions. When a reduced speed limit is in effect a yellow flashing light on top of the sign is activated and a reduced speed message is displayed The study showed that vehic le speeds were reduced by 0.47 to 0.75 mph for every 1 mph reduction in posted speed. In Seattle, Washington VSL s have been ins talled recently on a stretch of I 5 from B oeing access road to I 90. The project began in 2009 with the installation of fifteen new overhead sign bridges T he system was activated in August 2010. The overhead signs feature individual displays for each lane and warn of approaching lane closures and traffic congestion. The project is designed to reduce th e number of collisions and collision related congestion. The displayed speed limit ranges from 40 mph to 60 mph and is based on speed and volume data The speed limit is enforced by the Washington State Patrol. There has yet to be a formal assessment of t he effectiveness of the system (WSDOT 2010 ). 2.1.2 Implementation in Europe According to Hines (2002), numerous VSL systems have been implemented in Europe an countries. Based on European case studies he reported that VSLs can stabilize traffic flow in co ngestion and thus decrease the probability of crashes. The following provides an overview of VSL implementation in Europe. A VSL system was implemented along a n 18 km (11 mi le ) section of Autobahn 9 near Munich, Germany, in the 1970s. The system was origina lly implemented to improve safety, but the effects of the VSL system on other key parameters were also evaluated. The system displays speeds based on three control strategies: incident detection, harmonization, and weather conditions. Boice et al. (2006) i nvesti gated the effects of the system on key parameters around

PAGE 20

20 bottleneck formation based on one day data along the site. It was found that once a bottleneck had formed there was an 11% reduction in flow in the northbound direction and a 6% reduction in flow in the southbound direction. Capacity values were provided by lane and they were compared to the Highway Capacity Manual (HCM, 2000), and the German Handbuch fr die Bemessung von Strassenverkehrsanlagen (HBS, 2002). The capacity values for the median lane were consistent with both the HCM and the HBS values. The capacity value for the middle lane was consistent with the HBS but slightly lower than the HCM. The shoulder lane capacity was consistently lower than both manuals. It was concluded that there was no improvement in the capacity values over recognized standards. In the Netherland s a VSL system was installed along the A16 motorway near Breda in 1991. This system was designed to improve driving safety during fog conditions. The syst em has signs every 0.4 0.5 miles over 7.4 miles, 20 visibility sensors and automatic incident detection. The speed limit was reduced to 80 kilometers per hour ( km/h ) (50 mph ) from 100 km/h (62 mph ) if visibility dropped below 140 meters and was reduced to 60 km/h (37 mph ) from 100 km/h (62 mph ) if visibility dropped below 70 meters When an incident was detected, a speed limit of 50 km/h (31 mph ) was posted on the first sign upstream and 70 km/h (43 mph ) on the second sign upstream (Robinson, 2000). The results of an evaluation (Zarean et. al, 1999) showed that drivers reduced their mean speeds by about 8 10 km/h (5 6 mph ) during fog conditions No information could be found on the current status of the system, but it was operational in 2000. Another VSL system was ins talled in the Netherland s along a 20 km (12 mi) rural section of the A2 motorway between Amsterdam and Utrecht in 1992 (Robinson, 2002) The system is designed to reduce the risk of shockwaves, crashes, and congestion. Variable message signs are spaced app roximately every one kilometer and loop detectors spaced every half kilometer. The

PAGE 21

21 pos ted speed limits are determined by a system control algorithm based on 1 minute averages of speed and volume across all lanes. If an incident is detected, a speed of 50 k m/h (31 mph ) is displayed. The evaluation results showed that the severity of shockwaves and speed in all lanes were reduced (Van de Hoogen and Smulders, 1994) The vehicle speed and speed deviation decreased leading to fewer short headways as well as redu ced severity of shockwaves. The study showed no positive effect on capacity or flow, but cited the safety benefits of traffic homogenization. S peed limits were adjusted in England in response to the level of congestion on the M25 motorway in 1995. The obje ctive of the system was to smooth traffic flow by reducing stop start driving. The 22.6 km long system has VSL stations spaced at 1 km intervals, loop detectors at 500 meter intervals and closed circuit television Using loop detectors measuring traffic d ens ity and speed, speed limits are lowered in incr ements as congestion increases. The speed limits are lowered from 70 mph to 60 mph when volume exceed s 1 650 veh/h/ln and lowered to 50 mph when volume exceeds 2,050 veh/h/ln. Results showed that traffic a ccidents decreased by 10 15% and there was a very high compliance with the VSL system (Robinson, 2000). The VSL system is still functioning today. Rm (1999) investigated the effects of weather controlled speed limits and signs on driver behavior on t he F innish E18 site in Finland. The study looked at two scenarios compared to a control case: one in the summer where the maximum speed limit is 120 km/h (75 mph ), and one in the winter where the maximum speed limit is 100 km/h (62 mph ). The control cases were during normal operating conditions in the summer and winter months. In the winter, during adverse road conditions the speed was lowered from 100 km/h (62 mph ) to 80 km/h (50 mph ). A 3.4 km/h (2.1 mph ) decrease in speeds was observed. It was noted that duri ng adverse conditions

PAGE 22

22 reducing speeds compared to the control case. It was concluded that the system is very beneficial for improving safety when drivers have a diff icult time perceiving adverse conditions. In the summer, results showed that the 85 th percentile speed was decreased more than the mean speed, essentially reducing high end speeds. Both winter and summer scenarios showed that VSLs decreased the mean speed and standard deviation of speeds and demonstrated traffic homogenization. This was an experimental site and no information could be found as to the current status of the system. VSL s have been implemented in Sweden at 20 locations. Lind (2006) looked at the impacts of weather controlled VSL s on the E6 motorway in Halland and the traffic controlled VSLs on the E6 in Mlndal, s outh of Gothen burg. The E6 in Mlndal is a low speed urban motorw ay with normal speed limit of 70 km/h (43 mph ). The VSL s in Mlnd al were implemented as advisory speed limits in 2004 and changed to enforceable speed limits in 2006. This was part of a study to determine how VSL s were perceived by motorists in both enforceable and advisory conditions. The speed limit for free flow cond itions was raised to 90 km/h (56 mph ). In dense traffic the speed is reduced in a stepwise manner. At 950 v eh /h/ln the speed is reduced to 70 km/h (43 mph ) and can be reduced to 50 or 30 km/h (31 and 17 mph ) depending on the density. Two thirds of intervie wed drivers indicated that they supported the VSL system and said that it made them more attentive as to changes in traffic conditions. The same proportion reported a less hectic driving scenario and reduction of queue lengths. When the advisory speed limi t was displayed crashes were reduced by 20% and when the enforceable speed limit was displayed crashes were reduced by 40%. The results showed an increase in average speed for all driving conditions and as much as a 40 km/h (25 mph ) increase in potential

PAGE 23

23 q ueue formation scenarios. The study concluded there was an improvement in driving behavior for congested conditions, and a homogenization of traffic. Papageorgiou et al. (2008) studied the impact of VSLs on traffic flow behavior (flow occupancy diagrams) t hrough simulation of a motorway in Europe. The displayed speed was based on a threshold control algorithm, with possible speed limits of 60 mph 50 mph and 40 mph The study showed that the 50 mph setting showed the most changes in traffic flow that could be used for improving traffic efficiency. The 40 mph setting was useful at high occupancies for displaying safe speeds, but not for improving traffic efficiency. The average occupancy was found to be higher when the VSL is implemented. The study conclude d that the effect on capacity was not clear. In summary, VSLs have been implemented in numerous areas throughout the United States, and are widespread throughout Europe. Most of the VSL systems in the US have been implemented to address adverse weather co nditions. Several of the European systems however have been implemented to smooth flow and reduce congestion related crashes. Several studies showed that mean speeds decrease when a VSL is implemented Several studies showed the speed standard deviation to decrease as well, and that decrease has been associated with safety benefits. From the literature review it was not clear whether evaluation studies examined speeds upstream of the bottleneck and the impacts of the VSL both in space and time. Speed drop has typically been evaluated in terms of whether the speed limit was effectively reduced, but it is not clear whether the average speeds and/or travel times have been evaluated for the duration of the peak period and considering the entire section typicall y affected by congestion. There has been little evidence to suggest that implementing VSLs has the potential to increase capacity. The systems using weather and road conditions to display VSLs have been

PAGE 24

24 shown to reduce crashes and homogenize traffic cond itions. Among active systems, the minimum speed limits provided in the US are typically between 40 mph and 50 mph while those in Europe typically vary between 60 km/h (37 mph ) and 80 km/h (50 mph ). It is also common in European systems to display a speed of 50 km/h (31 mph ) during a detected accident scenario. 2.2 VSL Simulation Simulation is a valuable tool for assessing the impact of changes in the transportation system and selecting optimal alternatives without actually implementing and testing them in the field. Several studies have been conducted to evaluate various VSL algorithms prior to their implementation. This section provides an overview of such studies and summarizes their findings. Hegyi et al. (2003) present a predictive model for coordinati on of VSL s to suppress shockwaves at highway bottlenecks. The objective of this control mechanism is to minimize the time a vehicle spends in the given network. The METANET model is used to simulate the network, but was modified to incorporate the effect o f speed limits into the calculation logic. METANET is a second order macroscopic traffic flow model. The controller predicts the evolution of the network based on the current state of the network and a control input. The algorithm bases speed increments t hrough real time calculations of traffic flow, density, and mean speed. Safety constraints are implemented into the model to prevent large speed limit fluctuations (e.g., 10 km/h ) The model was applied to a benchmark freeway segment consisting of two node s connecting one link. The study compared the use of continuous valued speed limits and discrete valued speed limits to a base scenario with no control. The results showed that in all control cases the coordination of speed limits eliminated the shockwave, and restored the volume exiting the section to capacity sooner.

PAGE 25

25 Hegyi et al. (2005) continued work on model predictive control through coordination of VSLs and ramp metering. The study compared the results of simulated ramp metering, and ramp metering wi th VSL s on a simple network. The results showed that when used in conjunction the total time spent in the system was lower and resulted in higher outflow. The decision of which method to use depends on the demand of the on ramp and the freeway. It is sugge sted that VSLs should be used if speed limits can limit the flow sufficiently, however if the flow becomes too large, ramp metering should be implemented. The authors suggest that integrated use of both technologies will produce more favorable results than the use of each technology by itself. Lin et al. (2004) presented two online algorithm s for VSL controls at highway wor k zones. average headway for vehicles to merge onto adjacent lanes. It consisted of two modules: one to compute the initial speed of each VSL sign, and the second re sponsible for updating the displayed spe ed on each VSL sign The algorithm computes the appropriate speeds starting on the link directly upstream of the work zone. The algorithm computes the target density and appropriate speed for that segment and works u pstream to calculate appropriate speed limits. The second VSL algorithm was aimed to maximize the total throughput from the work zone under some pre The model looks at projected queue lengths and changes the upstream speed control signs based on the optimization of a through put function. The simulation results by CORSIM indicated that VSL algorithms can increase work zone throughputs and reduce total vehicle delays. Moreover, when VSL was implemented, speed variances were lower than other n on controlled scenarios, altho ugh th Lee et al. (2004) used a real time crash prediction model integrated with the microscopic simulator PARAMICS t o assess the safety effects of VSL s on a 2.5 km stretch of a sample

PAGE 26

26 freeway segment. The algorithm fo r changing speeds was relatively simple. Three detector locations relay information to the controller which averages their values into one crash potential value. A crash threshold is predefined, and when the crash potential exceeds this threshold the speed limit for all three detector locations was set based on a set of criteria. When crash potential exceeded the threshold, the speed limits were reduced from the design speed limit (90 km/h ) based on the average speeds: reduced to 50 km/h km/h reduced to 60 km/h if km/h reduced to 70 km/h km/h and reduced to 80 km/h if average speed > 80 km/h The results found that reduction in speed limits can reduce avera ge total crash potential, and the greatest reduction in crash potential occurred at the location of high traffic turbulence such as a bottleneck. However, the reduction in speed limit also increased the travel time. Thus, there was a trade off between saf ety benefit s and syste m travel time increase. T he results were not based on real traffic data and many assumptions in the simulation were not calibrated to field conditions The authors speculated that this may account for the increase in travel time. Lee et al. (2006) continued work using the simulator PARAMICS in combination with the real time crash prediction model described earlier, to analyze the effect of VSL s on safety. Simulation results showed that the system obt ained the greatest safety benefit w hen speed changes were gradually introduced (5 mph every 10 minutes). It was also found that it is best to base the displayed speed on the average speed of detectors immediately upstream and immediately downstream of the VSL location. However, the study h as several limitations. First, it assumed that drivers would compl y with the speed limit. Second it ignored the potential of m after reducing speed).

PAGE 27

27 Mitra and Pant (2005) evaluated the impact of a VSL syste m on a freeway work zone using the model VISSIM. The authors considered t hree scenarios: base scenario (no work zone) reduce d speed on the work zone link and reduced speed with reduced lane width The displayed speed was only changed through the work zon e and only one value indicating lowered speed was displayed. Through analysis of the data, a process was carried out for developing an equation to calculate expected delays for a reduced speed through a work zone. The authors concluded that this equation c ould help determine the proper speed through a work zone without the use of repeated simulation. Abdel Aty et al. (2006) evaluated the safety effect s of VSL s on I 4 in Orlando, Flori da using PARAMICS. This was part of a series of papers which reported res earch related to the I 4 system. The algorithm not only investigated lowering speeds upstream of congestion, but also raising speeds limits after a congested area. The VSL signs were changed based on data from a detector directly associated with the sign. The study evaluated two speed regimes: low speed, and medium to high speed. The results found that there was a safety benefit in medium to high speed regions but not in low speed situations (congested situati ons). It was also shown that the greatest improv ement in safety was achieved by abruptly changing speeds (15 mph ) rather than gradually changing them. A travel time study was also conducted and showed a significant reduction in travel time through the segment. It was further recommended that decreasing speed limits before congestion and increasing them after congestion has positive impacts on safety and travel time. In a subsequent study, Abdel Aty et al. (2008) studied the effect s of VSL on reducing crash risk on I 4 at different volume l oading scenario s using PARAMICS. There were a total of 24 treatments in the experiment based on the extent of speed change, speed change distance, and

PAGE 28

28 speed change duration (5 to 10 minutes) The study investigated the benefits of reducing the speed (5 10 mph ) entering a congested area and increasing the speed (5 mph ) past the congested area. Crash risks were computed from a crash prediction model that was based on traffic parameters. The study found that VSL s could reduce the rear end and lane change crash risk at low volume conditions, especially when lowering the upstream speed limit by 5 mph and raising the downstream speed limit by 5 mph Again, VSLs were not found to be effective in reducing crash risk during congested conditions. Abdel Aty and Dhindsa (2007) also conducted a micro simulation study using PARAMICS in order to determine the impact that VSL s an d ramp metering would have on the safety of a 9 mile stretch of I 4 in Orlando. The study also investigated the impact of VSL s and ramp metering on operational parameters like speed and travel time. The speed limits were changed based on thresholds of 5 minute averages of travel speed, and the ALINEA feed back algorithm was used for the ramp metering. It was concluded that i mplementation of VSL can increase ave rage speeds and decrease speed variation in the network as well as improve the risk index It was also shown that the best implementation strategy is one where the speeds are incremented by 5 mph over a half mile. It was also shown that for safety improve ments, a scenario where only downstream speeds are increased, outperformed a scenario where upstream speeds are decreased and downstream speeds increased. A third conclusion drawn by the authors was that VSL and ramp metering are more effective when integ rated together When used in conjunction they showed shorter travel times and higher speeds than ramp metering or VSL alone. Jiang and Wu (2006) used a cellular automaton model and showed that using multiple speed limits (where the speed limits decrease gr adually from upstream to downstream ) can help

PAGE 29

29 remove traffic jams. For a single small jam the concept is that by altering the speeds appropriately one can decrease the inflow toward a jammed area and increase the outflow. This will eventually result in t he jam being dissipated. Their model was not based on field data. Allaby et al. (2007) evaluated the impact of a candidate VSL system on a n 8 km section of the eastbound Queen Elizabeth Way an urban freeway in Toronto, Canada. The study was conducted us ing the microscopic simulator PARAMICS combined with a categorical crash model developed by Lee ( 2003 ). The VSL algorithm used was based on a logic tree that uses threshold values for flow, occupancy, and average travel speed. The base speed used was 100 k m/h (62 mph ) and it could be reduced to 80 km/h (50 mph ) and 60 km/h (37 mph ). The signs were arranged so there was never an abrupt change of speed limits (10 km/h difference) between signs. Each VSL sign was linked to an adjacent loop detector, and each s ign operates individually. The results of the simulation showed that implementation of VSL signs could significantly improve safety, however the authors concluded that the use of VSL signs increased the travel time for all traffic scenarios considered. Pi ao and McDonald (2008) assessed the safety benefits of in vehicle VSL s on motorways using the microscopic simulation model AIMSUN. Traffic on UK motorway M6 with speed limit of 70 mph was simulated under different scenarios. VSL s were applied when the spee d difference between a queuing section and the upstream section was larger than 20 km/h (12.4 mph ) and were provided to drivers through in vehicle information. The simulation assumed that all vehicles were equipped with the in vehicle devices. The adjuste d speed limits could be 60 km/h (37 mph ) 70 km/h (43 mph ) 80 km/h (50 mph ) 90 km/h (56 mph ), or 100 km/h (62 mph ) The simulation results showed that VSL reduced speed diff erences creating homogenization reduced very small time headways, small time to collision (TTC) ev ents, and lane change

PAGE 30

30 frequency This in effect reduced crash potential. The author s also indicated that there were potential safety risks in using the in vehicle VSL compared with roadside VSL : large speed variation s in speed could occur because vehicle device. Papageorgio u et al. (2008) used a quantitative model to investigate the impact of VSL implementation on traffic flow. VSL s were incorporated into the general second order traffic flow model METAN ET as a control component. The study evaluated the system based on a no control case, coordinated ramp metering, VSL, and integrated scenario. The freeway was set up as a constrained discrete time optimal control problem and solved using a feasible direct ion algorithm. It was shown that VSL s can substantially improve the traffic flow efficiency of a stretch of roadway especially when combined with coordinated ramp metering. The study concluded that when the optimal solution is applied to real motorway traf fic, the solution will inevitably become non optimum due to uncertainties in the real traffic stream. The authors suggested that future research could use the optimal solution to develop a suitable feedback control strategy and update the solution in real time. Carlson et al. (2010) expanded on the work of Papageorgiou et al. (2008) by using a similar method, to explore the parallels between ramp metering and applying VSL upstream of a potential bottleneck or high volume merging situation. The METANET seco nd order macroscopic model was altered to allow the VSLs to be incorporated. The study showed that when applied upstream, the VSL can act similarly to ramp metering where the flow is held back on the mainstream rather than on the ramp. The traffic arriving at the bottleneck is temporarily reduced and the system delays propagation of the congestion. Four scenarios were evaluated: no control, VSL control, ramp metering, and integrated control. The VSL case decreased total time spent in the system (TTS) by 15. 3%, and when VSLs and ramp metering are used in conjunction

PAGE 31

31 the TTS was reduced by as much as 19.5%. The study concluded that traffic flow and capacity can be improved through VSL use by reducing the capacity drop at bottlenecks. However, if the VSL is app lied at under critical conditions without the potential for bottleneck mitigation, mean speed is lowered and flow efficiency is decreased. Popov et al. (2008) proposed a speed limit control approach to eliminate shockwaves based on a distributed controller design The METANET environment was used for the simulation. In this design, each VSL sign has its own controller, but they all use the same structure and parameters. The proposed method requires using the appropriate amount of upstream and downstream d ata. Different scenarios were presented where each controller uses data from as many as 5 downstream controllers and one upstream controller. The maximum speed limit was 120 km/h (75 mph ), and could be lowered in increments of 10 km/h to a minimum of 50 km /h (31 mph ). The authors showed that a simple, linear, static controller using immediate neighbor information successf ully resolves a shockwave. The control scenario when compared to a scenario without controllers reduced total time spent in the network b y 20%. Ghods et al. (2009) used METANET to investigate the use of ramp metering and VSL in order to reduce peak hour congestion. An adaptive genetic fuzzy contro l was used and was compared to the traditional ALINEA controller. Local density, local speeds and queue length of the on ramp were used as input data to develop the fuzzy controller. The fuzzy controller processes this input data and provides a corresponding metering rate and two VSL s. The idea behind fuzzy logic is to have a controller that re sembles human decision making. It can process imprecise input data to arrive at a definitive conclusion. Rather than having precise threshold values that determine the output values of the controller, approximate multi valued boundaries are used. This a llows for input data to have partial membership to a category as opposed to the

PAGE 32

32 membership options only. The study showed that the genetic fuzzy ramp metering and VSL control improved TTS by 15.3% In summary, much r esearch has been conducted on the potential benefits of VSLs through the use of simulation. One set of studies has used VSLs as a control mechanism similar to that employed in ramp metering. These studies concluded that VSLs can be used to suppress shockwa ves at bottlenecks by implementing the VSL upstream of a bottleneck. Those studies reported that VSLs were effective in reducing TTS in the network, and their effect was more beneficial when combined with ramp metering. Another set of studies investigated the use of VSLs in micro simulators (VISSIM, PARAMICS, AIMSUN) and evaluated the safety benefits of such systems. These studies generally conclude d that VSLs can improve safety, as they tend to reduce speed variability. 2.3 VSL Algorithms This section pr ovides more detailed information regarding various VSL alg orithms that have been developed. Different algorithms have been developed based on the purpose of the VSL. The first part of this section discusses VSL algorithms developed to mitigate congestion and improve safety, while the second part focuses on algorithms developed to address weather and other issues. 2.3.1 Congestion and Safety Related Algorithms The following three algorithms aim to mitigate shockwave s and are based on a combination of param eters : Along A2 b etween Amsterdam and Utrechtin, 1992 Netherlands (implemented) Based on 1 minute averages of speed and volume across all lanes 50 km/h if incident occurs Severity of shockwaves and speed in all lanes were reduced Detailed information regar ding location of signs and detectors was not provided

PAGE 33

33 METANET simulation 2003 ( not implemented) ases speed increments through real time calculations of traffic flow, density, and mean speed Uses rolling horizon values to continuously update the optimal solution Showed that during a developing shockwave the model predictive control created a scenario with less congestion and higher outflow METANET simulation 2008 ( not implemented) Used individual controller for each VSL sign using data from as many as 5 downstream controllers and one upstream controller Reduced speeds in 10 km/h increments from 120 km /h to as low as 50 km/h S howed that a simple, linear, static controller using immediate neighbor i nformation successfully eliminates a shockwave The following two algorithms are based on flow: M25 Motorway, 1995 England (implemented) When flow > 1650 veh/ h/ln: 70 mph to 60 mph. When flow > 2050 veh/h/ln: lowered to 50 mph Accidents decreased by 10 15%, very high compliance Detailed information on location of signs and detectors not provided On the E6 motorway in Mlndal, 2006 Sweden (implemented) Free fl ow = 90 km/h 950 veh/h/ln = 70 km/h Speed can be reduced as low as 50 to 30 km/h When speeds were advisory there was a 20% crash reduction observed. For enforceable speed limits the crash reduction improved to 40%. Other impacts included average speed increase, homogenization of traffic, and reduction in queue length. The following algorithm i s based on average occupancy thresholds : I 4 in Orlando, FL 2009 (implemented) Classifies traffic as either free, light or heavy depending on occupancy value <16% = free >16% and < 28% = light >28% = heavy

PAGE 34

34 Reduces speeds in 10 mph increments from 50 mph to 30 mph The following algorithm is based on average travel speeds: PARAMICS simulation 2004 ( not implemented) Each VSL has an associated loop detector located ad jacent to it Three signs are grouped together and data for these signs was averaged into one value If a crash potential threshold is reached the displayed speed is dropped at all signs using a set of criteria (all signs display the same speed) 50 km/h if a vg. km/h 60 km/h if 60 < avg. km/h 70 km/h if 70 < avg. km/h 80 km/h if avg. speed > 80 km/h Reduced average total crash potential, especially at the bottleneck, but increased the overall travel time This algorithm is base d on a combination of flow, occupancy, and average speed using a logic tree. PARAMICS simulation 2007 ( not implemented) Each VSL sign is linked to an adjacent detector that operates individually For low volumes (less than 1,600 vehicles per hour per lan e ( vphpl ) ) occupancy is used as part of the criterion for reducing speeds. For higher volumes (more than 1,600 vphpl) occupancy is not considered. Ultimately average speed determines the displayed speed. This algorithm does not address gradual speed limit reduction as drivers are approaching the bottleneck. The simulation results showed that VSL signs could improve safety but that the travel time for all traffic scenarios considered were increased. 2.3.2 Weather Related and Other Algorithms The followin g four algorithms were developed to address weather related issues (visibility, wind speed, precipitation severity, etc.) :

PAGE 35

35 Along A16 motorway near Breda, 1991 Netherland (implemented) 100 km/h (normal) 80 km/h if visibility <140 meters 60 km/h if visibilit y < 70 meters Mean speeds reduced by about 8 10 km/h during fog conditions 25 km, between Hammina and Kotka, 1997 Finland (implemented) 120 km/h for good road conditions 100 km/h for moderate road conditions 80 km/h for poor road conditions On the Finnis h E18 site, 1998 Finland (implemented) Lowered from 100 to 80 km/h in winter Lowered from 120 to 100 km/h in summer Decreased both the mean speed and the standard deviation of speed A long a 19 mile section of I 75 1993 Tennessee (implemented) 5 to 10 per cent reduction in speed no crashes due to fog after implementation In summary, there are a number of existing algorithms based on different performance measures. For algorithms involving congestion mitigation or shockwave dampening, VSL signs are almost a lways associated with downstream detectors to decrease flow entering a congested area. Algorithms based on weather or road condition parameters usually deal with VSLs associated with adjacent detectors. In both cases it is most common to gradually lower th e speed limit in increments of 5 or 10 mph Most algorithms also use a safety measure that prevents adjacent signs from having more than a 10 mph difference between them. In addition, nearly all systems use a mechanism to prevent hysteresis, or rapid fluct uation between displayed speeds. Some systems use minimum time durations, and others use reverse thresholds to avoid this event. 2.4 CORSIM Simulation Software CORSIM i s a microscopic simulation program used to simulate a variety of traffic situations. CO RTE allows the user to interface directly with the CORSIM simulation tool and implement a variety of algorithms that can bypass CORSIM standard algorithms RTEs

PAGE 36

36 have been used to run simulations with actual hardware in the loop, and to simulate othe r ITS tools that context of the TSIS shell is displayed in Figure 2 2 (McTrans, 2009). The TShell is where users can interface with the program and specify inputs to the simu lation. It sends the input data to the CORSIM Driver Component which coordinates with the CORSIM property pages. The driver communicates with the CORSIM server that calls a series of exported functions that drive the simulation loop. At the same time the s erver calls the specified RTE functions. The CORSIM simulation has a number of specified call points where it can export RTE functions. These points include an initialization at the beginning of the simulation, after the completion of a time step, or at th e completion of the simulation. There is a allows the RTE to send messages to the CORSIM driver, which can display these messages on the end user screen. Data struct ures can be accessed by the RTE through shared memory. This gives access to and allows the RTE to control aspects of the simulation. RTEs are a powerful tool within CORSIM that allow the user to simulate new technologies and ideas (McTrans, 2009). 2.5 Lite rature Review Summary It is clear VSLs are prevalent in Europe and are becoming more popular in the United States. It is also clear that simulation is a powerful tool for assessing new technologies that have the potential to improve traffic flow and safety With this growing interest in VSL s, the need to assess their benefits with simulation becomes evident. Creating a framework to simulate VSLs in CORSIM will provide a convenient way to test various algorithms and assess potential benefits before actually implementing these strategies on a roadway.

PAGE 37

37 Table 2 1. Orlando Florida I 4 VSL control t hresholds Occupancy for decreasing speed l imit (%) Occupancy for increasing speed l imit (%) Speed l imit (mph) Free f low < 16 < 12 50 Light c ongestion 16 28 12 25 40 Heav y c ongestion > 28 >25 30

PAGE 38

38 Figure 2 1. Decision path for determining the new posted speed of the trigger VSLs. [Adapted from A llaby P., Hellinga B., and Bullock M. 2007. Variable Speed Limits: Safety and Operational Impacts of a Candidate Control Strategy for Freeway Applications. IEEE Transactions on Intelligent Transportatio n System 8 ( 4 ) pp.671 680 Figure 3. ] Figure 2 2. Operation of CORSIM and RTE s [ Adapted from The McTrans Center., 2009. Run Time Extensio 3, Figure 1.]

PAGE 39

39 CHAPTER 3 METHODOLOGY The research wa s executed in the following four steps: study site selection and data assembly, development of the run time extension ( RTE ) implement ation/testing of variable speed limit ( VSL ) algorithms and analysis of simulation results. Each of these steps is described briefly in the following paragraphs. 3.1 Study Site Selection and Data Assembly A 1 3 mile secti on of Interstate 95 in Miami, Flori da was selected as the study site. This study site was selected based on the availability of data the existence of recurring congestion along the site and the interest of FDOT to evaluate whether VSL would be an effective strategy at this location The site experiences recurrent congestion at several locations along its length. A calibrated Corridor Simulation ( CORSIM ) simulation network of this stretch of roadway was available and wa s used as the bas is for simulat ing the selected VSL algorithms. Detail ed information regarding the study site is provided in Chapter 4 of the thesis. 3.2 Development of RTE A total of three algorithms which vary mainly in the type of input parameters used in decision making were tested From the literature reviewed it is d esirable to have algorithms based on volume average travel speed, occupancy, or a combination of the t hree. The first algorithm tested is based on occupancy thresholds. The second algorithm is based on volumes, and the third use s average travel speed as o ne of several parameters. D i fferent threshold values we re tested for each algorithm to determine the sensitivity of each algorithm to these thresholds. Variables associated with the sign spacing such as spacing and number of signs we re also tested. After t he algorithms were selected and the specific scenarios finalized, the RTE module wa s built. This involves writing C++ code for the RTE to interact with CORSIM. T here are

PAGE 40

40 separate RTE s for each VSL algorithm and each scenario. The RTE uses inputs from detec tors within the CORSIM simulation at every time step, and determine s whether a speed change needs to be implemented. The RTE then output s to the CORSIM simulation the updated speed limit for each VSL sign. Chapter 5 describes the RTEs developed in greater detail. 3.3 Testing and Analysis of Algorithms in CORSIM The thr ee identified algorithms we re implemented into the CORSIM simulation through the RTEs developed Specific scenarios we re created for each algorithm to test the threshold values of each algor ithm as well as the location of each detector and V SL sign. Each algorithm wa s run a sufficient number of times for each of the scenarios tested. Output processing wa s performed for each run to generate outputs in the form of comma separated files. Conta ined in these files are traffic volumes, speeds, travel times, and miles traveled The data were aggregated for each detector location and averaged over the total number of runs. The results from the output processing we re used to analyze the performance o f each algorithm. Each scenario within a given algorithm is compared to the other scenarios of that algorithm. The scenarios are compared based on average travel speed, total travel time vehicle miles traveled, and throughput They are compared based on link performance measures surrounding the VSL implementation as well as overall system values to obtain the optimum threshold values for the algorithm. Th e scenario with the best network performance wa s selected and compared to the other optimized algorith ms. Qualitative comparisons are also provided to assess the performance of each algorithm. Chapter 6 provides the results of the simulation and the comparisons outlined above, along with conclusions regarding the effectiveness of VSL and specific algorith ms.

PAGE 41

41 CHAPTER 4 STUDY SITE DESCRIPTI ON This chapter first describes the study site and then provides an overview of the Corridor Simulation ( CORSIM ) network used in the simulation. 4.1 Study Site Selection The selected study site is a 13 mile section of I 95 running through Miami, Florida from I 395 to Miami Gardens Dr. in the northbound direction. An aerial view of the section is shown in Figure 4 1. This section of I 95 has two High Occupancy Tolling (HOT) lanes as well as ramp metering The variable s peed limit ( VSL ) control is limited to the general purpose lanes and does not have a direct effect on the HOT lane operations. The analysis of results is confined the general purpose lanes. The Florida Department of Transportation ( FDOT ) is currently consi dering the implementation of VSL s along I 95 at th is location The roadway is already equipped with inductive loop detectors that can obtain speed, volume, and occupancy. The inductive loop detectors have ID numbers based on data from the Statewide Transpo rtation Engineering Warehouse for Archived Regional Data (STEWARD) database. The locations of the loop detectors along the study section are shown in Figure 4 2 The speed profile over the entire I 95 section being analyzed is displayed in Figure 4 3. Thes e speeds are averaged over a 15 minute period during the onset of congestion. In the simulation this time period corresponds to time period 4. As shown t here are two noticeable bottlenecks; one is located i mmediately before the exit to the turnpike and the other is at the entry to NW 103 rd street The focus o f this study is the bottleneck area just before the turnpike and the upstream area affected by it

PAGE 42

42 4.2 Calibrated CORSIM Simulation A CORSIM model has already been developed to replicate current tra ffic operations on th e stretch of I 95 ( FDOT, 2011) A screen shot of the network is shown in F igure 4 4 The simulation replicates the entire 13 mile stretch. However, the VSL implementation is only implemented at one bottleneck location. The data used i n the previous ly referenced study for calibration of the network was obtained from the STEWARD database The data represents traffic conditions on October 7, 2009 from 3:30 p.m. to 6:30 p.m Fifteen minute averages of these parameters wer e used to generat e volumes input into the software. The data con sisted of speed averages and volumes used to calibrate each 15 minute period. These t hree hours were selected to include the p.m. peak period and the associated congestion formation and dissipation. The netwo rk was calibrated to match field recorded volumes and speeds over each time period and it replicates both the ramp metering and HOT lane operations. The ramp metering uses a constant metering rate that is not demand sensitive and thus it is not expected to interact with the VSL algorithms in these simulations The HOT lanes are modeled as a separate parallel facility with interchanges at various access points. Calibration results for each time period are provided in Appendix A

PAGE 43

43 Figure 4 1. Section of I 95 being simulated by CORSIM

PAGE 44

44 Figure 4 2 Location of detectors a long the s ection of I 95 study s ite containing proposed VSL i mplementation 600921 NORTH OF NW 151 ST 600801 SOUTH OF NW 131 ST (After Merging) 600831 SOUTH OF NW 135 ST 600851 NORTH OF OPA L OCKA BLVD (After Merging) 600891 SOUTH OF NW 151 ST 600841 NORTH OF OPA LOCKA BLVD (Before Merging) 600711 SOUTH OF NW 111 ST 600781 NORTH OF NW 119 ST 600731 SOUTH OF NW 119 ST 600791 SOUTH OF NW 131 ST (Before Merging) 600701 NORTH OF NW 103 ST 600641 SOUTH OF NW 103 ST 6 00931 SOUTH OF TURNPIKE

PAGE 45

45 Figure 4 3. Speed p rofile of the I 95 s ection for t ime p eriod 4

PAGE 46

46 Figure 4 4 Screen shot of I 95 network r epresented in CORSIM

PAGE 47

47 CHAPTER 5 DEVELOPMENT OF THE R TE This chapter provides a description of the run time extension ( RTE ) development. First the variable speed limit ( VSL ) algorithms used in the simulat ion are identified. Then the options for sign location variations tested with each algorithm are outlined. The construction and framework of the RTE programs as well as an overview of how the programs are running are discussed in the last section. 5.1 Se lection of Algorithms Based on the literature review, a total of three algorithms are selected which represent the major types of algorithms that have been tested and implemented elsewhere The study selected algorithms that use different measures for tri ggering a speed limit change, to evaluate the impacts of different types of algorithms The occupancy based algorithm was selected as it is the one that is currently in use in Orlando, Florida (PBS&J, 2009). The volume based algorithm was selected because it is implemented on the M25 in England (Robinson, 2000) with very good overall results. The third algorithm selected is based on a combination of flow, occupancy, and average travel speed, and it is based on a study of a freeway in Toronto, Canada (Allaby et al., 2007). This algorithm was selected because it seemed a promising alternative to the other two; however this one has not been implemented in the field. Each algorithm has a range of threshold values that we re tested. To prevent a rapid fluctuation of speed limits, each algorithm has one set of thresholds for lowering the speed limit and another set of thresholds for raising them. Each algorithm functions similarly within the freeway system. An inductive loop detector is located at the bottleneck, a nd relays 1 minute averages of speed, occupancy, and volume to a VSL sign upstream of this location. When a particular threshold value is reached the speed limit is reduced at the associated VSL sign.

PAGE 48

48 Similarly, when a parameter drops below one of the reve rse thresholds the speed limit is allowed to increase back to a higher speed. The speed limit is only allowed to drop by one increment at a time. For instance if the current speed limit is 60 mph and a threshold is reached that notifies the sign to drop to 40 mph, the speed is only reduced to 50 mph. This prevents drastic changes in the speed limit that might create driver confusion. 5.1.1 Algorithm Based on Occupancy The algorithm based on occupancy has two sets of threshold values; one for the decreasin g of speed limits and one for the i ncreasing of speed limits The VSL sign is linked to downstream detector location 600931, and the average occupancy is calculated over all of the lanes. The traffic is classified as either free flow, light congestion, or heavy congestion. If the occupancy crosses a threshold line the speed limit is decreased by an increment of five miles per hour. Similarly the speed limit may increase back to its previous value but not more than 5 m ph at a time. This algorithm is based on the current operating algorithm of the I 4 system (PBS&J, 2009) The I 4 implementation evaluates the speed limit every 120 seconds, and this study evaluates the speed limit every 60 seconds. The first threshold scenario uses the same values as the I 4 sy stem. The n ext two threshold scenarios are generated based on findings from NCHRP R eport 3 87 (Elefteriadou et al. 2009) That report reported occupancy thresholds as a function of the probability of breakdown at merge junctions The three threshold scenar ios are shown in Table s 5 1, 5 2, and 5 3. 5.1.2 Algorithm Based on Flow The algorithm based on volumes also uses two threshold values; one for the decreasing of speed limits and one for the increasing of speed limits. The VSL sign is linked to downstream detector location 600931, and average volume is compute d in vehicles per hour per lane. When a volume drops below a specified threshold, the speed limit is decreased according to the threshold

PAGE 49

49 crossed. To return to the original speed the volume must cross a different threshold. This prevents a rapid fluctuation of speed limits without using minimum time duration. The first se t of threshold values were obtained from a study conducted on the M25 in England (Robinson, 2000), and are shown in Table 5 4. The sec ond set of thresholds are obtained from speed flow diagrams in the 2000 Highway Capacity Manual (HCM 2000) and are shown in Table 5 5. The thresholds are obtained by locating the volume of traffic where speeds drop for a given free flow speed, using the associated volume as the threshold point. 5.1.3 Algorithm Based on a Logic Tree including Flow, Occupancy, and Average Speed In this algorithm speed limits are determined based on a logic tree that includes flow, occupancy, and average travel speed. The de cision making logic is shown in Figure 5 1. The algorithm first takes into account flow data from loop detector location 600921 If the volume is less than or equal to 1 7 50 vphpl, the next step is to consider occupancy. If occupancy is less than or equal t o 16%, the maximum speed limit is posted. If the occupancy is greater than 16%, average speed determines which speed is displayed. Going back to the first step, if the volume is greater than 1 7 50 vphpl, the logic skips straight to the average speed calcula tion. The speed to be displayed is then sent to the appropriate VSL sign. The thresholds shown in the figure are placeholders. The actual thresholds f or volume and occupancy are the thresholds that display the best performance in the first two algorithms. The avera ge travel speed thresholds tested are 50 mph and 45 mph This algorithm is based on research conducted on a candidate VSL system in Toronto, Canada (Allaby, 2007) 5.2 Sign Location Variations Along with the different algorithms and thresholds tes ted, four diffe rent sign locations are also tested. Each scenario uses the same detector placement but varies in the location of VSL s ign. The first scenario uses one VSL sign spaced approximately one half mile from the

PAGE 50

50 bottleneck sour ce. The second scenar io uses one sign spaced approximately one mile from the bottleneck source. The third scenario uses two signs, with the first sign placed approximately one half mile from the bottleneck source. The second sign is placed approximately one half mile upstream from the first sign location. The fourth scenario uses two signs, with the first sign placed approximately one mile from the bottleneck source. The second sign is placed approximately one mile upstream of the first sign. In the case of two signs operating, the upstream sign always displays a speed limit 5 mph higher than the downstream sign. For example if the downstream sign is displaying 55 mph the upstream sign also displays 55 mph If the downstream sign displays 50 mph the upstream sign still display s 55 mph If the downstream sign displays 45 mph the upstream sign displays 50 mph The four sign location scenarios are shown visually in Figure s 5 2 through 5 5 5.3 Implementing the algorithms on t he Corridor Simulation ( CORSIM ) network requires a dynamic link library (DLL) that interfaces with the CORSIM simulation in real time. CORSIM allows this DLL to be imported through a n RTE interface. The interface allows the DLL to import and export variabl es internal to CORSIM. Three different DLLs are built one for each type of algorithm. The general structure of the program works similarly for each case, but the rules and thresholds for the speed change logic differ between each program. A flowchart dis playing the general logic of the program is displayed in Figure 5 6. Upon initialization of the simulation, the DLL program identifies where VSL signs have been specified, and what detectors are used to control the VSL operation. The links affected by the VSL sign are also identified. This allow s the speeds on the downstream links to be updated when a speed limit change occurs During the initialization period the point processing i nterval is

PAGE 51

51 also defined. This determines how data is aggregated from the ind uctive loop detectors. For this set of scenarios the point processing interval has been set to 60 seconds. After initialization is complete the DLL is accessed at the call point PREFRESIMVEHICLE. This is every time step (one second) during the sim ulation b efore vehicle movement take s place. First the program checks to determine whether the simulation is still in the initialization period. If so the program exits the function and this is reassessed at the next time step. If the simulation is not in the init ialization period the current speed limit is assessed based on average values relayed from the specified inductive loop detectors. If it is determined that a speed change is to occur, the free flow speed is updated on the link containing the VSL sign. At t he same time a message is displayed as the simulation runs to indicate a speed change has occurred An example of a speed change message during the CORSIM simulation with the occupancy algorithm is displayed in Figure 5 7. After the free flow speed has bee n updated on the link containing the VSL sign, the free flow speed is updated on the downstream links The free flow speed s at the downstream links are updated every 15 seconds and the time of the speed change depends on the free flow speed and distance b etween the VSL sign and the downstream link. This creates a rolling speed change so that all the downstream links are not updated simultaneously. This method mimics a real world scenario where the first vehicle observing a speed change represents a rolling speed change through the downstream links. Figure 5 9 shows how the speed change would propagate downstream for a sample speed limit change. In the diagram the speed limit drops from 55 to 50 mph starting at the VSL sign. The downstream link speeds are th en updated every 15 seconds based on the length of the link and the free flow speed

PAGE 52

52 Table 5 1. Occupancy thresholds for displayed speed limits (s cenario 1 ) Traffic Category Occupancy for decreasing speed l imit (%) Occupancy for increasing s peed l imit (% ) Speed l imit (mph) Free f low < 16 < 12 55 Light c ongestion 16 28 12 25 50 Heavy c ongestion > 28 >25 45 Table 5 2. Occupancy thresholds for displayed speed limits (s cenario 2) Traffic category Occupancy for decreasing speed l imit (%) Oc cupancy for increasing s peed l imit (%) Speed l imit (mph) Free f low < 10 < 8 55 Light c ongestion 10 30 8 27 50 Heavy c ongestion > 30 >2 45 Table 5 3. Occupancy thresholds for displayed speed limits (s cenario 3) Traffic category Occupancy for decreasing speed l imit (%) Occupancy for increasing s peed l imit (%) Speed l imit (mph) Free f low < 20 % < 17% 55 Light c ongestion 20 35 % 17 32% 50 Heavy c ongestion > 35 % >32% 45

PAGE 53

53 Table 5 4. Volume threshold s for displayed speed limits (s cenario 1) Flow for decreasing speed l imit (vphpl) Flow for increasing speed l imit (vphpl) Speed l imit (mph) < 1650 55 > 1650 < 1450 50 > 2050 <1850 45 Table 5 5. Volume threshold s for displayed speed limits (s cenario 2) Flow for decreasing s peed l imit (vphpl) Flow for increasing speed l imit (vphpl) Speed l imit (mph) < 1450 55 > 1450 < 1250 50 > 2000 <1800 45

PAGE 54

54 Figure 5 1. Decision tree logic for third a lgorithm

PAGE 55

55 Figure 5 2 Use of o ne s ign s p aced a pproximately o ne half m ile Figure 5 3 Use of o ne s ign s paced a pproximately o ne m ile.

PAGE 56

56 Figure 5 4 Use of t wo s igns s paced a pproximately o ne h alf m ile a part Figure 5 5. Use of t wo s igns s paced a pproximately o ne m ile a part

PAGE 57

57 Figure 5 6 Screen shot of text outpu tting to screen w hen speed c hanges

PAGE 58

58 Figure 5 7 Flowchart of general RTE logic

PAGE 59

59 Speed Limit 50 mph 55 mph A Seed Limit 50 mph 55 mph B Speed Limit 50 mph 55 mph C Figure 5 8 Time series of speed limit propagation downstream from VSL sign location. A) 0 seconds. B) 15 seconds. C) 30 seconds. D) 45 seconds. E) 60 seconds F) 75 sec onds

PAGE 60

60 Speed Limit 50 mph 55 mph D Speed Limit 50 mph 55 mph E Speed Limit 50 mph 55 mph F Figure 5 8. Cont inued

PAGE 61

61 CHAPTER 6 IMPLEMENTATION AND A NALYSIS OF ALGORITHM S IN CORSIM This chapter pr esents the implementation of the algorithms and results obta ined from the simulations Section 6.1 describes all the scenarios that are implemented and how the simulations are carried out. Section 6.2 examines the results of the no control scenario, and identifies the bottleneck source. Section 6.3 provides a comparison between the thresholds and sign positioning for each algorithm. Section 6.4 provides a comparative analys is between the best performing scenarios among the three algorithms. The chapter concludes with a summary of the findings. 6.1 Implementation of RTE Scenarios in CORSIM A different run time extension ( RTE ) is created for each algorithm and a total of 24 s cenarios are developed and tested. The description of each scenario is shown in Table 6 1. The RTEs are loaded into the network by adding a new tool to the tool configuration list. The CORSIM Driver is specified as the tool to run the simulation and the a ssociated extension is a trf This process is shown in Figure 6 1. The RTE dynamic link library ( DLL ) is then loaded and the associated call points are specified. For the RTE created the only call points are the INITIALIZE, and PREFRESIMVEHICLE call points. This means tha t the RTE will interface with CORSIM at these times. This is shown in F igure 6 2. Initially ten runs were conducted from the no control scenario to obtain the final number of runs needed The number of runs required for each scenario was based on the average speed of vehicles in the network in miles per hour. The standard deviation of the initial ten runs was 0.2865. The acceptable error was set to be 0.2 mph, and a confidence interval of 95% was used. The acceptable number of

PAGE 62

62 runs computed was approximately eight. Since originally ten runs were used and this value exceeded the m inimum of eight, tens runs are used for all scenarios. O utput processing is performed to provide averages over t he ten runs. The simulation generat es a comma separated value file that aggregates evaluation parameters by time period. On the network level the parameters include average travel spe ed, vehicle miles traveled, total travel time and throughput On the lin k level speed profile plots are created, displaying the average speeds over a 3 mile section of roadway upstream from the bottleneck location Each scenario consists of twelve time periods, fifteen minutes each. This is a total of 3 hours of simulation t ime for each scenario. Each simulation requires 3000 seconds of initialization period to properly load the network with vehicles. 6.2 Analysis of the No Control Scenario The speed profile of the no control s cenario is displayed in F igure 6 3. From the res ults it is clear that a bottleneck forms on link 159 161, beginning sometime between time period 3 and 4. During time period 3 the speed at this location has dropped to 35 mph, and at time period 4 it has dropped below 25 mph. The congestion moves upstream and a ffects links as far as link 147 148, which is over 2 miles upstream of the bottleneck source. Congestion does not dissipate until time period 12, which is approximately two hours after the initial breakdown. This represents a typical evening peak per iod with recurring congestion on the I 95 network, and an ideal scenario to test the selected variable speed limit ( VSL ) algorithms. 6.3 Analysis of VSL Algorithm Performance This section examines each algorithm individually based on network and link statistics. The speed profil e and throughput plots are observed for each of the scenarios tested The effectiveness of the different sign location scenarios is also evaluated.

PAGE 63

63 6.3.1 Analysis of the Occupancy Based Algorithm A total of twelv e occupancy based scenarios are simulated This include s three sets of thresholds and four different sign configurations. The network wide performance measures are shown in T able 6 2 Based on these, t he scenario that shows the greatest improvement in traffic conditions is using two signs spaced one half mile apart and using the threshold scenario 2 ( based on NCHRP Report 3 87) The scenario 2 thresholds are shown in Table 6 3. This scenario displays the greatest improvement in network average travel speed with a 3.38 % (1.34 mph) increase in aver age trave l speed. The total travel time is decreased by 3.35% (255 hours) which is also the greatest among the occupancy scenarios. The speed profile for this scenario compared to the no control scenario is shown in Figure 6 4 The figure shows that befo re the breakdown the VSL control scenario results in a slight reduction in speeds. This is because the algorithm is reducing speeds below the free flow speed. During congested conditions the speed difference is negligible between the no control and control scenarios. However, s tarting at time period 5 a noticeable improvement in speed is observed starting at the bottleneck source and progressing upstream This trend continues, moving upstream and increasing average speeds by as much a s 16 mph Th ere is minimal improvement observed at the bottleneck but significant improvements can be observed in the upstream links. It appears that t he speed of the shockwave has been reduced, reducing the effects of congestion on upstream links. It als o appears that congestion does not extend as far upstream as in the no control scenario. The length of queue is shortened by approximately 1/3 of a mile. To evaluate the throughput of the section, scaled cumulative curves were constructed to provide the th roughput at key locations around the bottleneck. The scaled cumulative departures are used because they can show much more clearly the difference s between the control and no control case The scaled cumulative departures are obtained by subtracting the ti me multiplied

PAGE 64

64 by a base flow rate from the cumulative departures. This relationship is shown in equation 6 1 below: s caled cum. departures = cum. d epartures base flow rate time period (6 1) T he scaled cum ulative departures are obtained at three links along the section The first area observed is at link 146 147. This is the farthest point upstream where the bottleneck has affe cted traffic conditi ons, and a base flow rate of 6 840 vehicles per hour was used as the scaling factor. The next area observed is at link 153 154. This is in the middle of the section affected by the bottleneck, and a base flow rat e of 6 280 vehicles per hour was used as the scaling factor. The final location observed is at link 159 161. This is at the entrance to the bottlen eck, and a base flow rate of 8 120 vehicles per hour w as used as the scaling factor. Figure 6 5 shows the throughput of this scenar io compared to the no control case The vertical axis represents the scaled cumulative departures (vehicles) and the horizontal axis represents the time period. The VSL control scenario shows increased throughput over the duration of the congestion, and a t every location affected by the bottleneck As shown, t he throug hput increases by as much as 92 vehicles for a given 15 minute period. 6.3.2 Analysis of the Volume Based Algorithm A total of eight volume based scenarios we re tested. This includes two set s of thresholds, and four different sign configurations. The total network wide performance measures are shown in Table 6 4 There is no single scenario that can be identified as having the best performance with respect to all measures. There are two scena rios that perform best in specific measures. The first one uses one sign spaced one mile from the bottleneck, and uses the threshold scenar io 1 (based on the thresholds implemented on the M25 freeway in England, shown in Table 6 5 ) This scenario displays the greatest improvement in network average travel speed with a 3.64 % (1.44 mph) increa se. This scenario also displays the greatest decrease in total travel time by 3.41%

PAGE 65

65 (260 hours). The speed profile for this scenario compared to the no control scenari o is shown in Figure 6 6 The second s cenario produce d comparable results to the previous one I t use s the same threshold scenario ( s cenario 1) but uses two signs spaced mile apart The scenario displays the third best results for improved average speed and total travel time with changes of 2.67% (1.06 mph) and 2.51% (191 hours) respectively. The speed profile for this scenario compared to the no contro l scenario is shown in Figure 6 7 Some differences can be observed in the speed p rofiles of these two scenarios. Starting with free flow conditions and during the beginning of congestion t he one sign scenario shows slightly reduced speeds while the two sign scenario shows a smoother transition of speeds and smaller deviation from the no control scenario At the onset of congestion both scenarios mirror the no control scenario almost identically. As time progresses both scenarios show improved speeds starting at the bottleneck and moving upstream. The first scenario shows consistent average speed improveme nts by as much as 13 mph The second scenario shows a similar pattern with speeds increasing by as much as 16 mph During the recovery phase the one sign scenario shows greater speed improvement than the two sign scenario, though both show an improvement o ver the no control scenario Other scenarios from this algorithm showed similar improvements in average speed increase, and travel time reduction. The throughput for these two scenarios is shown i n Figure 6 8 and Figure 6 9 Both scenarios showed improvem ents in throughput over the duration of congestion. This improved throughput accounts for the increased average travel speeds, and reduced travel times. This finding was observed in all the volume based scenarios. While both scenarios discussed display im proved throughput, the scenario using two signs spaced mile apart shows greater

PAGE 66

66 improvement in terms of throughput by displaying increases of as much as 88 vehicles over a 15 minute period. 6.3.3 Analysis of the Multiple Parameter Based Algorithm The pa r ameter thresholds used for the multiple parameter scenarios are determined from the best performing scenarios from the occupancy and volume based sce narios. Threshold scenario 2 consistently showed the greatest travel time reduction for the occupancy bas e d algorithm, and scenario 1 consistently showed the greatest travel time reduction in the volume based scenarios. The resulting threshold conditions for the multiple parameter algorithm are shown in Figure 6 10 A similar situation occurred as with the vol u me based scenarios, where it is difficult to determine a best case scenario. The scenario with two signs spaced mi le from the bottleneck displays the most improvement for average travel speed and total travel time with changes of 2.29% (0.91 mph) and 2 .3% ( 176 hours) respectively. The speed profile of this scenario compared to the no control sce nario is displayed in Figure 6 11 The scenario using one sign spaced mile from the bottleneck displays the second best results for average speed and total tr avel time with changes of 1.08% (0.43 mph) and 1.12% (85.7 hour s) respectively. The speed profile for this scenario compared to the no control scenario is shown in Figure 6 12 From the speed profile of both these scenarios it can be observed that the ini tial drop in speeds comes much earlier than the no control case. It appears the drop in speeds caused the traffic breakdown to occur sooner than it normally would have. Through the duration of congestion the control case shows mild improvement over the no control case. During recovery the speed profile looks nearly identical to the no control case, and at some points show lower average speeds.

PAGE 67

67 The throughput plots for both these scenarios are shown in Figure 6 13 and Figure 6 14 The scenario using two sign s spaced mile apart show s a noticeable increase in throughput by as much as 84 vehicles over a 15 minute period (Figure 6 13). T he scenario using one sign spaced mile from the bottleneck showed little improvement if any compared to the no control cas e (Figure 6 14) T he scenario with 1 sign spaced 1 mile from the bottleneck produced conditions worse than the no control scenario for every network parameter. Observing the speed profil e in Figure 6 15 the average speed is lower during every time period at the bottleneck location. Also through the duration of the simulation the congestion never dissipates The throughput for this scenario is shown in Figure 6 16 The throughput for this scenario is less than the no control scenario over the entire durati on of the bottleneck. 6.4 Summary of Analysis Overall the results show that implementation of VSL increased average speeds and decreased travel times during the simulation Al l but one scenario tested show improvement in both of these categories and the se improvements were clearly seen when evaluating the study section on a link by link basis. H owever the magnitude of the improvement when viewed as an average over the analysis period is very small ( a maximum of 3.6% improvement in average travel ). In ge neral, the throughput over the section affected by the bottleneck showed improvement over the no control scenario. This improved throughput explains the improvement in average travel speed and reduced travel time. In these simulations t he volume based algo rithm shows the best overall improvement. The mul tiple parameter algorithm shows the least improvement overall, and one of its scenario s shows worsening of conditions compared to the no control scenario. One of the possible explanations for why the volume based algorithm performs better than the occupancy based one is that occupancy remains relatively flat over a

PAGE 68

68 wide set of volumes; thus the volume based algorithm is quicker to trigger a speed limit drop upstream. With respect to the multiple parameter al gorithm, it appears that it results in breakdown earlier than in the no control case. Perhaps this is related to the particular thresholds used, and would not occur for higher thresholds. Note that this analysis has not sought to thoroughly evaluate and compare these algorithms. To completely asses s each algorithm and document their relative merits and preferred applications a full optimization of thresholds and sign positioning would have to be performed. A comparison can be made between the use of one sign and the use of two signs. Using only one sign creates a sharp transition between speeds and results in lower speeds before the onset of congestion. Using two signs creates a smoother transition between speeds and as a result displays less reduction i n speeds before the onset of congestion. Ho wever, the best sign spacing is not the same for every alg orithm. Each algorithm performs best with different sign spacing The key improvement that VSL seems to have is in the speeds upstream of the bottleneck, a nd in the queue length during congested conditions. It appears that e v aluating conditions directly at the bottleneck source will not show the full effect of the VSL system. While conditions are slightly improved at this location for most algorithms the m ore significant improvements occur upstream of the bottleneck source. Also, comparing the average speeds for the entire system shows minimal changes and masks the significant effects of the VSL upstream of the bottleneck. Evaluating effects and considerin g the entire time and space of the peak period shows benefits u p to three miles up stream of the bottleneck source. A verage speeds are significantly improved at those locations compared to the no control case, and queue lengths are significantly reduced Qu eue lengths for the occupancy based algorithm typically show reductions in the range of 0.25 to 0.35 miles. The

PAGE 69

69 volume based algorithm shows improvements in the range of 0.25 to 0.45 miles. The multiple parameter algorithm showed minimal reduction of queue but at some points displayed decreases as much as 0.2 miles. Overall network changes appear to be small when conditions are averaged through the entire system but when considered that this VSL implementation is only af fecting two to three miles of a 13 mile stretch, these small changes become more significant

PAGE 70

70 Table 6 1. Description of scenarios t ested Algorithm Threshold s cenario Sign l ocation No VSL c ontrol Occupancy b ased 1 One s ign half mile spacing Occupancy b ased 2 One s ign half mile spacing Occupancy b ased 3 One s ign half mile spacing Occupancy b ased 1 One s ign one mile spacing Occupancy b ased 2 One s ign one mile spacing Occupancy b ased 3 One s ign one mile spacing Occupancy b ased 1 Two s igns half mile spacing Oc cupancy b ased 2 Two s igns half mile spacing Occupancy b ased 3 Two s igns half mile spacing Occupancy b ased 1 Two s igns one mile spacing Occupancy b ased 2 Two s igns one mile spacing Occupancy b ased 3 Two s igns one mile spacing Volume b ase d 1 One s ign half mile spacing Volume b ased 2 One s ign half mile spacing Volume b ased 1 One s ign one mile spacing Volume b ased 2 One s ign one mile spacing Volume b ased 1 Two s igns half mile spacing Volume b ased 2 Two s igns half mile s pacing Volume b ased 1 Two s igns one mile spacing Volume b ased 2 Two s igns one mile spacing Multiple p arameter One s ign half mile spacing Multiple p arameter One s ign one mile spacing Multiple p arameter Two s igns half mile spacing Multiple p arameter Two s igns one mile spacing

PAGE 71

71 Table 6 2. Networ k performance measures for the o cc u pancy based scenario Algorithm # of signs Spacing (miles) Threshold s cenario Average s peed % Change from no c ontrol Vehicle miles t raveled (veh) % Ch ange from no c ontrol Total travel t ime (hours) % Change from no c ontrol Total t hroughput (veh) % Change from no c ontrol No c ontrol 39.57 0.00 299905.2 0.00 7624.55 0.00 24530 .0 0 .00 Occupancy 1 1/2 1 40.13 1.41 299987.3 0.03 7514.82 1.44 24521.6 0.03 Occupancy 1 1/2 2 40.39 2.05 299803.8 0.03 7475.28 1.96 24533.8 0.0 2 Occupancy 1 1/2 3 40.60 2.60 299640.6 0.09 7425.28 2.61 24509.6 0.08 Occupancy 1 1 1 40.15 1.46 299785.2 0.04 7512.60 1.47 24519.6 0.04 Occupancy 1 1 2 40.82 3.14 299440. 8 0.15 7380.71 3.20 24475 .0 0.22 Occupancy 1 1 3 40.11 1.35 299888.7 0.01 7523.84 1.32 24533.8 0.0 2 Occupancy 2 1/2 1 40.37 2.02 299782.7 0.04 7471.69 2.00 24508.4 0.08 Occupancy 2 1/2 2 40.91 3.38 300043.3 0.05 7369.21 3.35 24544.4 0.0 6 Occup ancy 2 1/2 3 40.33 1.90 299602.4 0.10 7471.87 2.00 24528.1 0.0 1 Occupancy 2 1 1 40.82 3.14 299440.8 0.15 7380.71 3.20 19578.2 20.1 9 Occupancy 2 1 2 39.89 0.79 299789.3 0.04 7570.62 0.71 24501.6 0.1 2 Occupancy 2 1 3 40.03 1.15 299710.1 0.07 753 5.28 1.17 24517.3 0.05

PAGE 72

72 Table 6 3 Occupancy thresholds for displayed speed limits (s cenario 2) Traffic category Occupancy for decreasing speed l imit (%) Occupancy for increasing s peed l imit (%) Speed l imit (mph) Free f low < 10 < 8 55 Light c on gestion 10 30 8 27 50 Heavy c ongestion > 30 >2 45

PAGE 73

73 Table 6 4. Network performance measures for the volume based algorithm Algorithm # of signs Spacing (miles) Threshold s cenario Average s peed % Change from no c ontrol Vehicle miles t raveled (v eh) % Change from no c ontrol Total travel t ime (hours) % Change from no c ontrol Total t hroughput (veh) % Change from no c ontrol No c ontrol 39.57 0.00 299905.2 0.00 7624.55 0.00 24530 .0 0.00 Volume 1 1/2 1 40.52 2.40 299650.9 0.08 7445.97 2.34 245 12.9 0.0 7 Volume 1 1/2 2 40.36 1.97 299636.7 0.09 7467.51 2.06 24497.2 0.13 Volume 1 1 1 41.01 3.64 300071.9 0.06 7364.17 3.41 24544.6 0.06 Volume 1 1 2 40.18 1.52 299547.2 0.12 7498.70 1.65 24488.6 0.17 Volume 2 1/2 1 40.63 2.67 300278.7 0.12 7433.17 2.51 24619.4 0.36 Volume 2 1/2 2 40.39 2.06 299521.1 0.13 7462.65 2.12 24492.1 0.15 Volume 2 1 1 40.86 3.25 299989.8 0.03 7388.90 3.09 24514.6 0.06 Volume 2 1 2 39.64 0.18 299799.6 0.04 7619.88 0.06 24535.9 0.02 Table 6 5. Network perf ormance measures for the multiple parameter based algorithm Algorithm # of signs Spacing (miles) Threshold Scenario Average Speed % Change from No Control Vehicle Miles Traveled (veh) % Change from No Control Total Travel Time (hours) % Change from No Con trol Total Throughput (veh) % Change from No Control No c ontrol 39.57 0.00 299905.2 0.00 7624.55 0.00 24530 .0 0.00 Multiple 1 1/2 1 40.00 1.08 299937.9 0.01 7538.87 1.12 24520.1 0. 04 Multiple 1 1 1 38.73 2.14 299544.0 0.12 7791.58 2.19 24459.5 0. 29 Multiple 2 1 /2 1 40.48 2.29 299763.0 0.05 7448.92 2.30 24486.0 0.18 Multiple 2 1 1 39.94 0.92 299433.1 0.16 7543.46 1.06 24436.6 0. 38

PAGE 74

74 Figure 6 1. Setting up the RTE tool configuration

PAGE 75

75 Figure 6 2. Specifying the call points of the RTE

PAGE 76

76 A B C Figure 6 3. Speed profile for no control scenario over time the 12 periods A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) T ime Period 10. K) Time Period 11. L) Time Period 12.

PAGE 77

77 D E F Figure 6 3. Continued

PAGE 78

78 G H I Figure 6 3. Continued

PAGE 79

79 J K L Figure 6 3. Continued

PAGE 80

80 A B C Figure 6 4. Speed profile for the occupancy based algorithm using two signs spaced mile apart (threshold scenario 2) compared to the no control scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) T ime Period 11. L) Time Period 12.

PAGE 81

81 D E F Figure 6 4. Continued

PAGE 82

82 G H I Figure 6 4. Continued

PAGE 83

83 J K L Figure 6 4. Continued

PAGE 84

84 A B C Figure 6 5 Throughput for the occupancy based algorithm using two signs spaced mile apart (threshold scenario 2) compared to the no control scenario A) A t entry to bottleneck area (link 146 147). B) A t the middle of the st udy section (link 153 154). C) At the upstream end of congestion (link159 161). -20 0 20 40 60 80 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottleneck Area No Control VSL Control 0 100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput in the Middle of the Study Section No Control VSL Control 0 100 200 300 400 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 85

85 A B C Figure 6 6 Speed profile for the volume based algorithm using one sign spaced 1 mile from the bottleneck source (thresh old scenario 1 ) compared to the no control scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) Time Period 11. L) T ime Period 12.

PAGE 86

86 D E F Figure 6 6 Continued

PAGE 87

87 G H I Figure 6 6 Continued

PAGE 88

88 J K L Figure 6 6 Continued

PAGE 89

89 A B C Figure 6 7 Speed profile for the volume based algorithm using two signs spaced 1/2 mile apart (threshold scenario 1) comp ared to the no control scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) Time Period 11. L) Time Period 12.

PAGE 90

90 D E F Figure 6 7 Continued

PAGE 91

91 G H I Figure 6 7 Continued

PAGE 92

92 J K L Figure 6 7 Continued

PAGE 93

93 A B C Figure 6 8 Throughput for the volume based algorithm using one sign spaced 1 mile from the bottleneck source (threshold scenario 1 ) compared to the no control scenario A) At ent ry to the bottleneck area (link 146 147). B) At the middl e of the study section (link 153 154). C) At the upstream end of congestion (link159 161). -20 0 20 40 60 80 100 120 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottleneck Area No Control VSL Control -100 0 100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput in the Middle of the Study Section No Control VSL Control 0 100 200 300 400 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 94

94 A B C Figure 6 9 Throughput for the volume based algorithm using two signs spaced 1/2 mile apart (threshold scenario 1 ) compared to the no control scenario A) At entry to the bottleneck area (link 146 147). B) At the middle of the stud y section (link 153 154). C) At the upstream end of congestion (link159 161). -20 0 20 40 60 80 100 120 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottlneck Area No Control VSL Control 0 100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput in the Middle of the Study Section No Control VSL Control 0 100 200 300 400 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 95

95 Figure 6 10 Logic tree thresholds used for the multiple parameter algorithm

PAGE 96

96 A B C Figure 6 11 Speed profile for the multiple parameter -based algorithm using two signs spaced 1/2 mile apart compared to the no control scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) Time Pe riod 11. L) Time Period 12.

PAGE 97

97 D E F Figure 6 11 Continued

PAGE 98

98 G H I Figure 6 11 Continued

PAGE 99

99 J K L Figure 6 11 Continued

PAGE 100

100 A B C Figure 6 12 Speed profile for the multiple parameter based algorithm using one sign spaced 1/2 mile from the bottleneck source compared to the no control scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) Time Period 11 L) Time Period 12.

PAGE 101

101 D E F Figure 6 12 Continued

PAGE 102

102 G H I Figure 6 12 Continued

PAGE 103

103 J K L Figure 6 12 Continued

PAGE 104

104 A B C Figure 6 13 Throughput for the multiple parameter based algorithm using two signs spaced 1/2 mile apart compare d to the no control scenario A) At entry to bottleneck area (link 146 147). B) At the middle of the study sect ion (link 153 154). C) At the upstream end of congest ion (link159 161). -10 0 10 20 30 40 50 60 70 80 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottlneck Area No Control VSL Control -100 0 100 200 300 400 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput in the Middle of the Study Section No Control VSL Control 0 50 100 150 200 250 300 350 400 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 105

105 A B C Figure 6 14 Throughput for the multiple parameter based algorithm using one sign spaced 1/2 mile from the bottleneck source compared to the no control scenario A) At entry to the bottleneck area (link 146 147). B) At the middle of the study section (link 153 154). C) At the upstream end of congestion (link159 161). -10 0 10 20 30 40 50 60 70 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottlneck Area No Control VSL Control -100 0 100 200 300 400 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput in the Middle of the Study Section No Control VSL Control 0 50 100 150 200 250 300 350 400 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 106

106 A B C Figure 6 15 Speed profile for the multiple parameter based algorithm using one sign spaced 1 mile from the bottleneck source compared to the no contr ol scenario A) Time period 1. B) Time Period 2. C) Time period 3. D) Time period 4. E) Time period 5. F) Time Period 6. G) Time Period 7. H) Time Period 8. I) Time Period 9. J) Time Period 10. K) Time Period 11. L) Time Period 12.

PAGE 107

107 D E F Figure 6 15 Continued

PAGE 108

108 G H I Fig ure 6 15 Continued

PAGE 109

109 J K L Figure 6 15 Continued

PAGE 110

110 A B C Figure 6 16 Throughput for the multiple parameter based algorithm using one sign spaced 1 mile from the bottleneck source compared to the no control scenario A ) At entry to bottleneck area (link 146 147). B) At the middle of the study section (link 153 154). C) At the upstream end of congestion (link159 161). -10 0 10 20 30 40 50 60 70 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at Entry to Bottlneck Area No Control VSL Control -50 0 50 100 150 200 250 300 350 400 450 500 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Middle of the Study Section No Control VSL Control -100 -50 0 50 100 150 200 250 300 350 1 2 3 4 5 6 7 8 9 10 11 12 Scaled Cumulative Departures (veh) Time Period Throughput at the Upstream End of Congestion No Control VSL Control

PAGE 111

111 CHAPTER 7 SUMMARY AND CONCLUSI ONS Variable speed limits ( VSL s) have been used upstream of bottlenecks with recurring congestion as a way to dampen the shockwave produced once congest ion starts. However, the exact effects of VSL on traffic flow are not well understood, and the literature provides conflicting conclusions with respect to the effectiveness of VSL installations in increasing overall speeds and throughput. Simulation is a v ery effective tool in evaluating alternatives under completely controlled conditions which cannot be achieved in the field. It is also very effective in providing a comprehensive picture of traffic operations in time and space. This study creates a test bed for evaluating different VSL control algorithms and implementations through the use of the micro simulator Corridor Simulation ( CORSIM ) A run time extension ( RTE ) tool is built and used to simulat e VSL along a freeway s ection to allow the evaluation of multiple algorithms under varying conditions. The RTE is created to model the implementation of a VSL system using C++ code. The dynamic link library ( DLL ) created interfaces with the CORSIM simulation at specifi ed c all points. Detector data are averaged every 1 minute and the speed limit is evaluated based on given threshold s Three different algorithms are tested along with a number of threshold scenarios, and sign positioning One algorithm uses thresholds based on occupancy, the second one uses thresholds based on volumes, and the last one uses thresholds based on volume, occupancy, and average travel speed. T he following were concluded : Through all 24 of the scen arios tested, all but one show an improvement in te rms of average travel speed and total travel time. The magnitude of the change is relatively small when averaged through the entire network. The t hroughput was found to increase for most of the VSL scenario s tested by a maximum of 30 to 90 vehicles over a given 15 minute time period

PAGE 112

112 Of the scenarios tested the volume based algorithm d isplays the largest improvement in terms of increased average speed, and reduced travel time T he effect of the VSL may not be immediately seen if one examines conditions o nly at the bottleneck. The area u pstream of the bottleneck shows much greater traffic improvements than the bottleneck itself. Improper selection of thresholds or sign positioning can cause traffic conditions to become worse than if no VSL control is used. This is shown by the multiple parameter algorithm (scenario 2), where conditions worsened for every evaluation parameter and time period There was no consistent trend in traffic conditions as a function of the number and location of speed limit signs. Th e best sign positioning was found to be highly dependent on the type of algorithm and specific thresholds selected. The results and conclusions made by this study assume a level of compliance from motorists. In order for these results to parallel real w orld implementation, there must be the same level of compliance. This emphasizes the need for enforcement of the speed limits when they are implemented in the field The inclusion of a user interface in the normal installation of CORSIM should be considere d for future releases of the software. Recommendations for a permanent interface include the following: Develop an alternative to the rolling speed limit approach. This could be accomplished by assigning speed limit changes on a vehicle basis, rather than on a link basis. Include input s creens for each type of VSL algorithm It is recommended to have volume occupancy, and speed based VSL algorithms available as options. Allow the user to specify threshold values to be associated with speed limit changes fo r each type of algorithm simulated Incorporate an interface to place VSL signs at a desired location, and link it to specified detectors using a visual interface. Recommendations for future research include the following: This study did not attempt to tho roughly evaluate and compare the three VSL algorithms, nor to obtain optimal thresholds for each type of algorihtm. An optimization type study could be p erform ed to obtain optimal thresholds, sign locations, and detector locations.

PAGE 113

113 The study could be ext ended to consider additional bottlenecks along the I 95 corridor. Research considering ramp metering in conjunction with VSL would be useful in determining the mechanism through which these two tools would interact, and developing guidelines for optimizin g their joint operation F ield testing of the algorithms tested here and a compar ison between simulation and field results would be useful in validating, and as necessary enhancing the simulation algorithm.

PAGE 114

114 APPENDIX A CALIBRATION RESULTS FOR I 95 NETWORK This appendix provides the calibration results of the I 95 CORSIM simulation. The network was calibrated based on 11 detector locations. Th is appendix provides the volume calibration information over all 12 time periods, followed by the speed calibration data over all 12 time periods. Figure A 1. Volume calibration for time p eriod 1 Figure A 2. Volume calibration for time p eriod 2

PAGE 115

115 Figure A 3. Volume calibration for time period 3 Figure A 4. Volume calibration for time period 4

PAGE 116

116 Figure A 5. Volum e calibration for time period 5 Figure A 6. Volume calibration for time period 6

PAGE 117

117 Figure A 7. Volume calibration for time period 7 Figure A 8. Volume calibration for time period 8

PAGE 118

118 Figure A 9. Volume calibration for time period 9 Figure A 10. Vol ume calibration for time period 10

PAGE 119

119 Figure A 11. Volume calibration for time period 11 Figure A 12. Volume calibration for time p eriod 1 2

PAGE 120

120 Figure A 13. Speed calibration for time p eriod 1 Figure A 14. Speed calibration for time period 2

PAGE 121

121 Figure A 15. Speed calibration for time period 3 Figure A 16. Speed calibration for time period 4

PAGE 122

122 Figure A 17. Speed calibration for time period 5 Figure A 18. Speed calibration for time period 6

PAGE 123

123 Figure A 19. Speed calibration for time period 7 Figure A 20. Speed calibration for time period 8

PAGE 124

124 Figure A 21. Speed calibration for time period 9 Figure A 22. Speed calibration for time p eriod 1 0

PAGE 125

125 Figure A 23. Speed calibration for time p eriod 1 1 Figure A 24. Speed calibration for time p eriod 1 2

PAGE 126

126 APPEN DIX B SAMPLE SOURCE CODE F OR RTE This appendix gives the source code for the RTE of one scenario. This is the occupancy based algorithm using 2 signs spaced mile apart using threshold scenario 1. The header file that imports and redefines variables is sh own first, followed by the main DLL source code. ifndef _FRESIM_H_ #define _FRESIM_H_ #define DLL_IMPORT extern "C" __declspec( dllimport ) #define DLL_EXPORT extern "C" __declspec( dllexport ) #define IMXNOD 8999 DLL_IMPORT struct { int YINIT; } GLR091; #define yinit GLR091.YINIT DLL_IMPORT s truct { float ZCLOCK; } PRI306; #define zclock P RI306.ZCLOCK DLL_IMPORT struct { int TTLFLK; } PRI215; #define ttlflk PRI215.TTLFLK DLL_IMPORT struct { int NFMAP[IMXNOD]; } PRI075; #define nfmap PRI075.NFMAP DLL_IMPORT struct { int DPPINT; } GDET01; #define dppint GDET01.DPPINT DLL_IMPORT float* FRESIM_DETECTORS_mp_ZFDOCC; #define zf docc FRESIM_DETECTORS_mp_ZFDOCC DLL_IMPORT float* FRESIM_DETECTORS_mp_ZFDSPD; #define zfdspd FRESIM_DETECTORS_mp_ZFDSPD DLL_IMPORT float* FRESIM_DETECTORS_mp_ZFDVOL; #define zfdvol FRESIM_DETECTORS_mp_ZFDVOL DLL_IMPORT float* FRESIM_LINKS_mp_ZFFLOW; #de fine zfflow FRESIM_LINKS_mp_ZFFLOW DLL_IMPORT int* FRESIM_LINKS_mp_DWNODC; #define dwnodc FRESIM_LINKS_mp_DWNODC

PAGE 127

127 DLL_IMPORT int* FRESIM_LINKS_mp_UPNODC; #define upnodc FRESIM_LINKS_mp_UPNODC DLL_IMPORT int* FRESIM_LINKS_mp_DWNODE; #define dwnode FRESIM_ LINKS_mp_DWNODE DLL_IMPORT int* FRESIM_LINKS_mp_UPNODE; #define upnode FRESIM_LINKS_mp_UPNODE DLL_IMPORT int* FRESIM_DETECTORS_mp_DETLK; #define detlk FRESIM_DETECTORS_mp_DETLK DLL_IMPORT int* GLOBAL_LINKS_mp_INMAP; #define inmap GLOBAL_LINKS_mp_INMAP DLL_IMPORT int* FRESIM_LINKS_mp_FDETID; #define fdetid FRESIM_LINKS_mp_FDETID DLL_IMPORT int* FRESIM_DETECTORS_mp_NDETLK; #define ndetlk FRESIM_DETECTORS_mp_NDETLK #endif //_FRESIM_H_

PAGE 128

128 // This is the main DLL file. #include "stdafx.h" #include "VSL.h" #include "CORWin.h" #include "fresim.h" // Declare Functions and Global Variables int getDet(int link, int config); int getLinks(int up, int down); void updateSpeed(int link, int d1, int d2, int d3, int d4, int d5, int d6, int d7, int d8, int d 9); void moveSpeeds(int lnk, float spd[], int maxTime); void upSign(int lnk, float spd[], int maxTime); int L159L161, L161L165, L165L166, L166L167, L158L159, L258L158, L154L155, L155L156, L156L157, L157L158; int d1, d2, d3, d4, d5; float evalTime = 60.00 0000, horizonTime = 60.000000; float speed[10] = {73, 73, 73, 73, 73, 73, 73, 73, 73, 73}; float upSpeed[10] = {73, 73, 73, 73, 73, 73, 73, 73, 73, 73}; DLL_EXPORT void _stdcall vsl_Initialize(){ // Reassigning link numbers to match CORSIM Internal Link Numbering to // User Defined Link Numbering getLinks(upstream node, downstream node) L154L155 = getLinks(154, 155); L155L156 = getLinks(155, 156); L156L157 = getLinks(156, 157); L157L158 = getLinks(157, 158); L158L159 = getLinks(158, 1 59); L159L161 = getLinks(159, 161); L161L165 = getLinks(161, 165); // Assigning detector numbering getDet(link, lane) d1 = getDet(L159L161, 1); d2 = getDet(L159L161, 2); d3 = getDet(L159L161, 3); d4 = getDet(L159L161, 4); d5 = getDe t(L159L161, 5); // Setting the Point processing Interval dppint = 60;

PAGE 129

129 } DLL_EXPORT void _stdcall vsl_PreFreesimVehicle() { if (yinit != 1){ if (zclock == evalTime){ evalTime = evalTime + 60; updateSpeed(L158L159, d1, d2, d3, d4, d5, 9 999, 0, 0, 0); } if (zclock == horizonTime){ horizonTime = horizonTime + 15; moveSpeeds(L158L159, speed, 9); zfflow[L159L161] = speed[1] + 7; zfflow[L161L165] = speed[2] + 7; upSign(L154L155, upSpeed, 9); zfflow[L154L155] = upSpeed[0 ]; zfflow[L155L156] = upSpeed[1]; zfflow[L156L157] = upSpeed[2]; zfflow[L157L158] = upSpeed[4]; } } } int getLinks(int up, int down){ int dnode = 0; int unode = 0; int LinkID = 0; for (int index = 0; index < ttlflk; index++){ int dnode = d wnode[index]; int unode = upnode[index]; if (dnode < 7000) {dnode = nfmap[dnode 1];} if (unode < 7000) {unode = nfmap[unode 1];} if (up == unode && down == dnode){ LinkID = index; } } return LinkID; } int getDet(int link, int config){ int detectorID = 0;

PAGE 130

130 if (fdetid[link] != 0){ if (config == 1){ detectorID = fdetid[link] 1; } if (config == 2){ detectorID = (fdetid[link] 1) + 2; } if (config == 3){ detectorID = (fdetid[link] 1) + 4; } if (config == 4){ detectorID = (fdetid[link] 1) + 6; } if (config == 5){ detectorID = (fdetid[link] 1) + 8; } if (config == 6){ detectorID = (fdetid[link] 1) + 10; } if (config == 7){ detectorID = (fdetid[link] 1) + 12; } if (config == 8){ detectorID = (fdetid[link] 1) + 14; } if (config == 9){ detectorID = (fdetid[link] 1) + 16; } } return detectorID; } void updateSpeed(int link, int d1, int d2, int d3, int d4, int d5, int d6, int d7, int d8, int d9){ int n = 0; int d etNum = 0; float average = 0; if (d9 == 9999){ detNum = 8;} if (d8 == 9999){ detNum = 7;} if (d7 == 9999){ detNum = 6;} if (d6 == 9999){ detNum = 5;} if (d5 == 9999){ detNum = 4;} if (d4 == 9999){ detNum = 3;} if (d3 == 9999){ detNum = 2;} if (d2 == 9999){ detNum = 1;}

PAGE 131

131 switch (detNum) { case 0: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3] + zfdocc[d4] + zfdocc[d5] + zfdocc[d6] + zfdocc[d7] + zfdocc[d8] + zfdocc[d9])/9; break; case 8: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3] + zfdocc[d4] + zfdocc[d5] + zfdocc[d6] + zfdocc[d7] + zfdocc[d8])/8; break; case 7: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3] + zfdocc[d4] + zfdocc[d5] + zfdocc[d6] + zfdocc[d7])/7; break; case 6: average = (zfdocc[d1] + zfdocc[d2] + zfdo cc[d3] + zfdocc[d4] + zfdocc[d5] + zfdocc[d6])/6; break; case 5: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3] + zfdocc[d4] + zfdocc[d5])/5; break; case 4: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3] + zfdocc[d4])/4; break; case 3: average = (zfdocc[d1] + zfdocc[d2] + zfdocc[d3])/3; break; case 2: average = (zfdocc[d1] + zfdocc[d2])/2; break; case 1: average = zfdocc[d1]; break; } If (zfflow[link] >73) n = 0; else if (zfflow[link] >62 && zfflow[link] < 70) n = 1; e lse if (zfflow[link] < 60) n = 2; switch (n) { case 0: if(average > 16){ zfflow[link] = 66; char text[132]; sprintf_s( text, "Speed Limit has been reduced to 50 MPH, with Average Occupancy:%f \ n", average); OutputString( text, 132, 2 0 ); } else { zfflow[link] = 73; } break;

PAGE 132

132 case 1: if(average <= 12){ zfflow[link] = 73; char text[132]; sprintf_s( text, "Speed Limit has been Increased back to 55 MPH, with Average Occupancy:%f \ n", average); OutputString( text, 132, 2, 0 ); } if (average > 12){ if(average > 28){ zfflow[link] = 58; char text[132]; sprintf_s( text, "Speed Limit has been Reduced to 45 MPH, with Average Occupancy:%f \ n", average); OutputStr ing( text, 132, 2, 0 ); } if (average <= 28) { zfflow[link] = 66; } } break; case 2: if (average > 25){ zfflow[link] = 58; } else{ zfflow[link] = 66; char text[132]; sprintf_s( text, "Speed Limit has been Increased back to 50 MPH, with Average Occupancy:%f \ n", average); OutputString( text, 132, 2, 0 ); } break; } } void moveSpeeds(int lnk, float spd[], int maxTime){ for (int i = maxTime; i > 0; i -){ spd[i] = spd[i 1]; }

PAGE 133

133 spd[0] = zfflow[lnk]; } void upSign(int lnk, float spd[], int maxTime){ for (int i = maxTime; i > 0; i -){ spd[i] = spd[i 1]; } if (speed[0] < 60){ spd[0] = speed[0] + 7; } else { spd[0] = 73; } }

PAGE 134

134 APPENDIX C SPEED PROFILES FOR S CENARIOS TESTED Th is appendix displays the speed profiles over all 12 time periods for each scenario tested. A B C D E F Figure C 1. Speed profile for occupancy based algorithm using 1 sign spaced mile from the bottleneck source (threshold scenario 1) compare d to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 135

135 G H I J K L Figure C 1. Continued

PAGE 136

136 A B C D E F Figure C 2 Speed profile for occupancy based algorithm using 1 sign spaced1/2 mile from the bottleneck source (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 137

137 G H I J K L Figure C 2 Continued

PAGE 138

138 A B C D E F Figure C 3 Speed profile for occupancy based algorithm using 1 sign spaced1/2 mile from the bottleneck source (threshold scenario 3) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Ti me period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 139

139 G H I J K L Figure C 3 Continued

PAGE 140

140 A B C D E F Figure C 4 Speed profile for occupancy based algorithm using 1 sign spaced1 mile from the bottleneck source (threshold scenario 1) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) T ime period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 141

141 G H I J K L Figu re C 4 Continued

PAGE 142

142 A B C D E F Figure C 5 Speed profile for occupancy based algorithm using 1 sign spaced1 mile from the b ottleneck source (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 143

143 G H I J K L Figure C 5 Continued

PAGE 144

144 A B C D E F Figure C 6 Speed profile for occupancy based algorithm using 1 sign spaced1 mile from the bottleneck source (threshold scenario 3) compared to th e no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 145

145 G H I J K L Figure C 6 Continued

PAGE 146

146 A B C D E F Figure C 7 Speed profile for occupancy based algorithm using 2 signs spaced1/2 mile apart (threshold scenario 1) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 147

147 G H I J K L Figure C 7 Continued

PAGE 148

148 A B C D E F Figure C 8 Spee d profile for occupancy based algorithm using 2 signs spaced1/2 mile apart (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 149

149 G H I J K L Figure C 8 Continued

PAGE 150

150 A B C D E F Figure C 9 Speed profile for occupancy based algorithm using 2 signs spaced1/2 mile apa rt (threshold scenario 3) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 1 1. L) Time period 12.

PAGE 151

151 G H I J K L Figure C 9 Continued

PAGE 152

152 A B C D E F Figure C 10 Speed profile for occupancy based algorithm using 2 signs spaced1 mile apart (threshold scenario 1) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 153

153 G H I J K L Figure C 10 Continued

PAGE 154

154 A B C D E F Figure C 1 1 Speed profile for occupancy based algorithm using 2 signs spaced 1 mile apart (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time perio d 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 155

155 G H I J K L Figure C 1 1 Continued

PAGE 156

156 A B C D E F Figure C 1 2 Speed profile for occupancy based algorithm using 2 signs spaced 1 mile apart (threshold scenario 3 ) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time peri od 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 157

157 G H I J K L Figure C 1 2 Continued

PAGE 158

158 A B C D E F Figure C 13. Speed profile for volume based algorithm using 1 sign spaced1/2 mile from the bottleneck source (threshol d scenario 1) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time p eriod 12.

PAGE 159

159 G H I J K L Figure C 13. Continued

PAGE 160

160 A B C D E F Figure C 1 4 Speed profile for volume based algorithm using 1 sign spaced1/2 mile from the bottlen eck source (threshold scenario 2 ) compared to the no control case. A) Ti me period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 161

161 G H I J K L Figure C 1 4 Co ntinued

PAGE 162

162 A B C D E F Figure C 15. Speed profile for volume based algorithm using 1 sign spaced 1 mile from the bottleneck source (threshold scenario 1) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Ti me Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 163

163 G H I J K L Figure C 15. Continued

PAGE 164

164 A B C D E F Figure C 1 6 Speed pr ofile for volume based algorithm using 1 sign spaced 1 mile from the bottleneck source (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Ti me period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 165

165 G H I J K L Figure C 1 6 Continued

PAGE 166

166 A B C D E F Figure C 1 7 Speed profile for volume based algorithm using 2 sign s spaced1 /2 mile apart (threshold scenario 1 ) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Ti me period 11. L) Time period 12.

PAGE 167

167 G H I J K L Figure C 1 7 Continued

PAGE 168

168 A B C D E F Figure C 1 8 Speed profile for volume based algorithm using 2 sign spaced 1/2 mile apart (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 169

169 G H I J K L Figure C 1 8 Continued

PAGE 170

170 A B C D E F Figure C 1 9 Speed profile for volume based algorit hm using 2 sign s spaced 1 mile apart (threshold scenario 1 ) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Ti me period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 171

171 G H I J K L Figure C 1 9 Continued

PAGE 172

172 A B C D E F Figure C 20 Speed profile for volume b ased algorithm using 2 sign spaced 1 mile apart (threshold scenario 2) compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 173

173 G H I J K L Figure C 20 Continued

PAGE 174

174 A B C D E F Figure C 21. Speed profile for multiple parameter based algorithm using 1 sign spaced1/2 m ile from the bottleneck source compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 1 2.

PAGE 175

175 G H I J K L Figure C 21 Continued

PAGE 176

176 A B C D E F Figure C 22. Speed profile for multiple parameter based algorithm using 1 sign spaced 1 mile from the bottlene ck source compared to the no control case. A) Time period 1. B) Tim e period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 177

177 G H I J K L Figure C 22 Continued

PAGE 178

178 A B C D E F Figure C 23. Speed profile for multiple parameter -based algorithm using 2 sign s spaced1/2 mile apart compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time period 10. K) Time period 11. L) Time period 12.

PAGE 179

179 G H I J K L Figure C 23 Continued

PAGE 180

180 A B C D E F Figure C 24. Speed profile for multiple parameter -based algorithm usi ng 2 sign s spaced 1 mile from the bottlene ck source compared to the no control case. A) Time period 1. B) Time period 2. C) Time period 3. D) Time Period 4. E) Time period 5. F) Time period 6. G) Time period 7. H) Time period 8. I) Time period 9. J) Time p eriod 10. K) Time period 11. L) Time period 12.

PAGE 181

181 G H I J K L Figure C 24 Continued

PAGE 182

182 LIST OF REFERENCES Abdel Aty M., Dilmore, J., and Hsia, L. 2006. Applying Variable Speed Limits and the Potential for Crash Migration. Transportation Rese arch Record 1953, pp. 21 30. Abdel Aty, M., Dilmore, J., Dhindsa, A. 2006. Evaluation of Variable Speed Limits for Real Time Freeway Safety Improvement. Accident Analysis and Prevention 38, pp. 335 345. Abdel Aty, M., and Dhindsa, A. 2007. Coordinated Us e of Variable Speed Limits and Ramp Metering for Improving T raff ic Safety on Congested F reeways. In: 86th Annual Meeting of the Transportation Research Board Washington, D.C ., Paper No. 07 0008 Abdel Aty M., Cunningham R. J., Gayah V. V., and Hsia L. 20 08. Dynamic Variable Speed Limit Strategies for Real Time Crash Risk Reduction on Freeways. Transportation Research Record 2078, pp. 108 116. Allaby P ., Hellinga B., and Bullock M., 2007. Var iable Speed Limits: Safety and O perational Impacts of a Candidate Control St rategy for Freeway Applications. IEEE Transactions on Intelligent T ransport ation System 8 ( 4 ) pp. 671 680 Boice, S., Bertini, R. L., Ahn, S., and Bogenberger, K. 2006. Dynamics of Variable Speed Limit System Surrounding Bottleneck on German A utobah n. Transportation Research Record 1978, pp. 149 159. Buddemeyer, J., Young, R.K., Dorsey Spitz, B., 2010. Rural Variable Speed Lim it System for Southeast Wyoming. In: 89 th Annual Meeting of the Transportati on Research Board, Washington D.C., Paper No 10 2444. Carlson, R.C., Papamichail, I., Papageorgiou, M., Messmer, A., 2010. Variable Speed Limits as a Mainli ne Metering Device. In: 89 th Annual Meeting of the Transportati on Research Board, Washington D.C., Paper No. 10 1529. Florida Department of Tr ansportation (FDOT)., 2011. Managed Lane Operations Adjusted Time of Day Pricing vs. Near Real Time Dynamic Pricing. Tallahassee, FL. Elefteriadou, L., Brilon, W., Jacobson, L., 2011. Proactive Ramp Management Under the Threat of Freeway Flow Breakdown NCHRP Report 3 87. Ghods, A.H., Rahimi Kian, A., Tabibi, M. 2009. Adaptive Freeway Ramp Metering and Variable Speed Limit Co ntrol: A Genetic Fu zzy Approach. IEEE ITS Magazine 1 (2), pp. 27 36. Goodwin, L.C., 2003. Best Practices for Road Weather Managem ent, Version 2.0 Miretek S ystems, Inc., Falls Church, VA. Haas R., Carter M., Perry E., Trombly J., Bedsole E., Margiotta R. 2009. iFlorida Model Dep loyment Final Evaluation Report, Washington, D.C. FHWA HOP 08 050.

PAGE 183

183 Hegyi A., De Schutter B., and Hellen doorn J., 2003. MPC based Optimal Coordination of Variable Speed Limits to Suppress Shock Waves in Freeway T ra c. In: Proceedings of the 2003 American Control Conferen ce, Denver, Colorado. pp. 4083 4088. Hegyi A., De Schutter B., and Hellendoorn J. 20 05. Model predictive control for optimal coordination of ramp metering and variable speed limits. Transportation R esearch Part C, 13 (3), 185 209. Highway Capacity Manual (HCM), 2000. Transportation Research Board. Washington, DC, ISBN: 0 309 06681 6. Hine s, M. 2002. Judicial Enfor cement of Variable Speed Limits. NCHRP Legal Research Digest 47. Jiang R., Wu Q., 2006. Suppressing Jams by Multiple Speed L imit s: a Simulation Study Based on Cellular Automaton Model. In: Proceedings of the 84 th Annual Meeting of the Transportation Rese arch Board, Washington DC. Lee, C., Hellinga B., Saccomanno, F., 2004. Assessing Safety Benefits of Variable Speed Limits. Transportation Research Record 1897 pp. 183 190. Lee, C., Hellinga, B. Saccomanno, F., 2006. Evaluation o f Variable Speed Limits to Improve T ra c Safety. T ran sportation Research Part C 14 (200) pp. 213 228. Lin, P., Kang, K., and Chang, G. 2004. Exploring the Effectiveness of Variable Speed Limit Controls on Highway Work Zone Operations. Intellige nt Transportation Systems 8, pp.1 14. Lind, G., 2006. Weather and Traffic Controlled Variable Speed Limits in Sweden. Movea trafikkonsu lt, Stockholm, Sweden. McLawhorn, N., 2003. Variable Speed Limit Signs for Winter Weather Wisconsi n Department of Transportation. The McTrans Center., 2009. Run Time E Gainesville, Florida. Mitra A. and Pant P. D. 2005. A Framework to Evaluate the Impact of Variable Speed Limit Systems on Work Zone Traffic Operation Using VISSIM. In: Proceedings of the Institute of Transpo rtation Engineers District 6 Annual Meeting, Kalispell, Montana. Papageorgiou, M., Kosmatopoulos, E., and Papami chail I., 2008. Effects of Variable S peed Limits on Motorway Traffic Flow Transportation Research Record 2047, pp. 37 48. PBS&J., 2009. I 4 V ariable Speed Limit Effectiveness Study. Prepared for the Florida Department of Transportation, District 5.

PAGE 184

184 Piao J. and McDonald M. 2008. Safety Impacts of Variable Speed Limits A Simulation Study. In: Proceedings of the 11th International IEEE Conferen ce on Inte lligent Transportation Systems, Beijing, China. Popov, A., Babuska, R., Hegyi, A. and Werner, H., 2008. Distributed Controller Design for Dynamic Speed Limit C ontrol Against Shock Waves on Freeways. In: Proceedings of the 17th IFAC World Congres s, Seoul, Korea, pp 14060 14065. Placer J., 2001. Fuzzy Variable Speed Limit Device Mo dification and Testing Phase II. Arizona Departme nt of Transportation. Rm, P.,1999. Effects of Weather Controlled Variable Speed Limits and Warning Signs on Driver B eh avior. Transportation Research Record 1689, pp. 53 59. Robinson M. D. 2002. Safety Applications of ITS in Rural Areas U.S. Department of Transportat ion, Washington D.C. Robinson, M. D. 2000. Examples of Variable Speed A pplication. In: Proceedings of th e 79 th Annual Meeting of the Transportation Rese arch Board, Washington, D.C. Sisiopiku V., 2001. Variable Speed Control: Technologies and Practices. In: Proceedings of the 80 th Annual Meeting of the Transportation Re search Board Washington, D.C. Steel, P., R.V. McGregor, A.A. Guebert and T.M. McGuire. 2005. Application of Variable Speed Limits along the Trans Canada Highway in Banff National Park. In: Proceedings of the Annual Conference of the Transport ation Association of Canada, Calgary, Alber ta. Ulfarsson G., Shankar, V., Vu, P., Mannering, F., Boyle, L., Morse, M., 2001. TravelAid Washington State Transportation Center, University of Washington, Washington. Van den Hoogen E ., and Smulders S 1994. Control b y Variable Speed Signs: Results of the Dutch E xperiment. In: Seven th International Conference on Road Traffi c Monitoring and Control. London, UK, pp.145 149. Washington State Department of Transportation. I 90 Two Way Transit and HOV Operations Variable Speed Limit Signs. [online] Available at: < http://www.wsdot.wa.gov/Projects/I90/TwoWayTransit/vsl.htm > [Accessed September 13, 2010]. Zarean, M., Pisano, P., Dirnberger, K., and Robinson, M., 1999. Varia ble Speed Limit Systems : The State Of The Practice. In: Proceedings of the 1999 Rural Advanced Technology & Transportation Systems Conference, Flagstaff, AZ.

PAGE 185

185 BIOGRAPHICAL SKETCH Clark Letter was born in Melbourne, Florida in 1985. In 2009 he received hi s Bachelor of Science in civil engineering from the Univer sity of Florida, Gainesville, Florida In 2011 he earned his Master of Science in civil engineering from the University of Florida. During his graduate studies he was a research assistant for his advisor, Dr. Lily Elefteriadou.