2021

Xu Zhang (xuzhang_uk@uky.edu), Kentucky Transportation Cabinet
Reginald Souleyrette, Kentucky Transportation Cabinet
Eric Green, Kentucky Transportation Cabinet
Teng Wang, Kentucky Transportation Center
Mei Chen, Kentucky Transportation Center
Paul Ross, Kentucky Transportation Center

Traffic incidents remain all-too-common events.  They negatively impact the safety of the traveling public and emergency responders and cause significant traffic delays.  Congestion associated with incidents can instigate secondary crashes, exacerbating safety risks and economic costs.  Traffic incident management (TIM) provides an effective approach for managing highway incidents and reducing their occurrence and impacts.  The paper discusses the establishment of and methods of calculation for five TIM performance measures that are used by the Kentucky Transportation Cabinet (KYTC) to improve incident response.  The measures are: Roadway Clearance Time, Incident Clearance Time, Secondary Crashes, First Responder Vehicle Crashes, and Commercial Motor Vehicle Crashes.  Ongoing tracking and analysis of these metrics aid the KYTC in its efforts to comprehensively evaluate its TIM program and make continuous improvements.  As part of this effort, a fully interactive TIM dashboard is developed using the Microsoft Power BI platform.  Dashboard users can apply various spatial and temporal filters to identify trends at the state, district, county, and agency level.  The dashboard also supports dynamic visualizations such as time-series plots and choropleth maps.  With the TIM dashboard in place, KYTC personnel, as well as staff at other transportation agencies, can identify the strengths and weakness of their incident management strategies and revise practices accordingly.

Sakib Khan (sakibk@clemson.edu), Clemson University
Anthony Patire, University of California, Berkeley

Transportation agencies monitor freeway performance using various measures such as VMT (Vehicle Miles Traveled), VHD (Vehicle Hours of Delay), and VHT (Vehicle Hours Traveled). Public transportation agencies typically rely on point detector data to estimate these freeway performance measures. Point detectors, such as inductive loops cannot capture the travel time for a corridor, which can lead to inaccurate performance measure estimation. This research develops a hybrid method, which estimates of freeway performance measures using a mix of probe data from third-parties and data from traditional point detectors. Using a simulated I-210 model, the overall framework using multiple data sources is evaluated, and compared with the traditional point detector-based estimation method. In the traditional method, point speeds are estimated with the flow and occupancy values using the g-factors. Data from 5% of the total vehicle are used to generate the third-party vendor provided travel time data. The analysis is conducted for multiple scenarios, including peak and off-peak periods. Findings suggest that fusing data from both third-party vendors and point detectors can help estimate performance measures better, compared to the traditional method, in scenarios that have noticeable traffic demand on freeways.

Yao Cheng (chengyao09@gmail.com), University of Maryland, College Park
Gang-Len Chang, University of Maryland

To alleviate the concerns of the overflowing on-ramp queues to impede local streets, a standard practice of ramp metering control either at the local or coordinated level is to restrain its function when a series of preset conditions are identified by on-ramp queue detectors. such a trade-off between potential ramp queue spillback and restraint of a metering control’s function often may neither effectively mitigate on-ramp waving contributed bottlenecks, nor convince arterial users and local traffic agencies of the essentiality of ramp metering operations. Hence, this study presents an arterial-friendly local ramp metering system (named AF-ramp) that can yield the target metering rate to best the freeway conditions but not to spill ramp queues back to the surface street, by concurrently optimizing the signal plans for those intersections sending turning flows to the ramp. Such a system is developed for time-of-day control at this stage, and can serve as the base module for extending to real-time on-line control, or multi-ramp coordinated operations. AF-ramp model with its functions to optimize the arterial signals concurrently with the ramp metering rate can best use the capacity of local intersections and prevent any gridlock by overflows from the on-ramp queue spillback or the arterial turning traffic. With extensive simulation experiments, the evaluation results confirmed the developed AF-ramp model’s effectiveness with respect to concurrently improving traffic conditions on both the freeway and its neighboring arterial links. This study has also developed a framework to extend the AF-ramp model from the time-of-day control to real-time mode.

Hao Liu, University of California, Berkeley
Xiao-Yun Lu, University of California, Berkeley
Zhitong Huang, Leidos, Inc.
Steven Shladover, University of California, Berkeley
Soomin Woo, University of California, Berkeley

Variable speed advisory (VSA) has the potential to mitigate stop-and-go waves and traffic oscillations that could decrease capacity. This research aimed to identify the impact of vehicle connectivity and users’ acceptance on the performance of a feedback variable speed advisory system. It adopted a simulation approach to conduct scenario analyses to estimate the performance of the VSA algorithm under various connected manually driven vehicle (CMDV) market penetrations. The analysis results indicate that the VSA control could have substantial effects on the freeway corridor when the CMDV market penetration was 10 to 40 percent. With the advisory speed, vehicle fuel efficiency increased 2–6 percent. These results suggest that the speed adaptation of a few connected drivers could change the traffic flow pattern, leading to more energy-efficient traffic flow. The performance of the VSA algorithm was not sensitive to small changes in the driver compliance level. However, the traffic performance changed significantly when the CMDV drivers fully comply with the VSA. The full compliance brought about 2 to 3 percent extra benefit on vehicle fuel efficiency. If the VSA algorithm could generate advisory speed based on the predicted traffic conditions rather than the current conditions, the effects of CMDV should become further increased. The research findings are essential for understanding the mechanism of traffic pattern change due to the implementation of VSA in a partially connected traffic stream. Those findings are important for refining the existing VSA algorithms and improving the CV systems to support the VSA application.

Zhe Xiao (zestshaw@gmail.com), Southeast University
Xiaoyu "Sky" Guo, Texas A&M University, College Station
Xiucheng Guo, Southeast University
Yi Li, Southeast University

Many studies have modeled and evaluated the impact of Cooperative Adaptive Cruise Control (CACC) on multilane highways, but they assumed under a uniform speed limit policy. Existing research have revealed the traffic performs differently under a differentiated per-lane speed limit (DPLSL) policy. Such DPLSL policy, is frequently combined with restraints on heavy vehicles’ entry to inner lane(s), and is commonly used in countries, e.g., China and Thailand. Whether the benefits of CACC still remain under a DPLSL policy has not been explored yet. Hence, in this study, a cellular automation model was developed incorporating CACC equipped passenger cars, CACC non-equipped passenger cars and heavy vehicles on a two-way eight-lane highway with the DPLSL policy. The result indicates that CACC technology can increase the lane throughput by up to 65% as the CACC market penetration rate (MPR) rises. However, the rise is not significant until the CACC MPR exceeds 40%. At the same time, heavy vehicles induce a 5%~17% reduction on the throughput of all the lanes even under high CACC MPRs (60% and 80%). Furthermore, although the multilane highway system is under a high CACC MPR, because of the DPLSL policy, a decrease in throughput is still experienced in those lanes with different maximum and minimum speed limits from its left adjacent lane, regardless of heavy vehicle percentage. The study is helpful for policy makers to further prepare for the CACC’s prevalence in the forthcoming years.

Adrian Sandt, University of Central Florida
Haitham Al-Deek, University of Central Florida

Limited access facility wrong-way driving (WWD) crashes are typically more severe than other crashes.  Deploying advanced WWD countermeasures, such as rectangular flashing beacon (RFB) and light-emitting diode (LED) technologies, at exit ramps can reduce WWD crashes, injuries, and fatalities.  No previous research has developed a methodology to quantify the potential fatality and injury savings due to future countermeasure deployments.  This paper developed such a methodology and applied it to the Florida’s Turnpike Enterprise (FTE) toll road network.  From 2011–2016, there were 53 FTE WWD crashes, resulting in 16 fatalities and annual injury costs of $37 million.  87% of these crashes occurred during nighttime.  RFB and LED life-cycle injury savings and costs were determined for all 216 FTE exits.  The total savings were $424 million for RFBs (benefit-cost ratio of 23.20) and $144 million for LEDs (benefit-cost ratio of 13.13).  Deploying countermeasures at the 103 exits with the highest benefit-cost ratios would provide 70% of the total possible savings by equipping 40% of the ramps.  For the same capital investment, RFBs provide more savings than LEDs.  Spending $1 million to deploy RFBs will provide similar savings as spending $3.4 million to deploy LEDs.  Evaluating the existing FTE RFB and LED ramps shows that RFBs are more effective at nighttime and can provide three times the savings of LEDs.  The results of this paper show the improved performance of RFBs over LEDs and provide an example that other agencies could follow to identify savings and cost-effectively deploy advanced WWD countermeasures.

Adrian Sandt, University of Central Florida
Haitham Al-Deek, University of Central Florida
Patrick Blue, University of Central Florida
John McCombs, University of Central Florida

Wrong-way driving (WWD) crashes can have significant impacts on freeway safety and operations.  Deploying Intelligent Transportation Systems (ITS) WWD countermeasures at freeway exit ramps can effectively reduce WWD crash risk (WWCR), but these countermeasures are expensive.  In this paper, a WWCR segment model and WWD countermeasures optimization algorithm are developed for all Florida limited access facilities (56 roadways with 1375 exits) to identify the optimal locations for ITS countermeasure deployment.  These were previously developed for specific toll-road networks within Florida, but never for a statewide network.  Multiple WWCR models were investigated, with the final Poisson model using four-exit segments and five years of WWD event data.  This model showed that more WWD events and higher crossing road traffic volumes increased WWCR, while certain interchange designs increased or decreased WWCR.  Sixty-three segments containing 169 exit ramps without ITS WWD countermeasures were identified as WWD hotspots; these ramps were compared to the 169 ramps with the highest WWCR selected by the optimization algorithm.  The algorithm selected 96 ramps not in the hotspots (improved resource utilization of 56.8%) and provided a 38.6% increase in WWCR reduction.  Comparing WWD detection and turnaround data from 31 sites with Rectangular Flashing Beacon ITS countermeasures to the optimization indicated a significant positive association between WWCR and turnaround percentage (higher WWCR at sites with lower turnaround percentage), verifying the accuracy of the optimization.  By showing the transferability, scalability, accuracy, and benefits of this approach, this paper can help agencies reduce WWD and improve freeway safety and operations.

Anas Abdulghani, University of Windsor
Chris Lee (cclee@uwindsor.ca), University of Windsor

This study develops a differential variable speed limit (DVSL) which assigns different speed limits for car and truck, and varies speed limits based on traffic conditions. The proposed DVSL algorithm changes speed limits in real time based on truck percentage and occupancy immediately upstream of the ramp, and the average speed of the control road sections upstream of the ramp. DVSL algorithm also considers spatial coordination of speeds, which gradually changes the speed limits in successive road sections upstream of the ramp when the severe congestion occurs. The study tested the impacts of DVSL and three other speed limit strategies on delay and safety for a section of the Gardiner Expressway in Toronto, Canada using the VISSIM traffic simulation model. The other strategies are 1) uniform speed limit (USL), 2) differential speed limit for car and truck (DSL), and 3) USL & DSL (U&D) – i.e., USL at low truck percentage and DSL at high truck percentage. It was found that DVSL showed the lowest delays for both car and truck among the four strategies. This is mainly because DVSL increased the spacing between vehicles in the right lane upstream of the on-ramp and facilitated vehicles’ merging into the mainline freeway. It was also found that DVSL showed the lowest likelihood of rear-end crash between the lead and following vehicles among the four strategies. This study demonstrates that the proposed DVSL algorithm can better control car and truck speeds to reduce delay and improve safety of car-truck mixed traffic flow on freeways.

John Hourdos, University of Minnesota, Twin Cities
Melissa Duhn, University of Minnesota, Twin Cities
Zhejun Liu, Facebook Inc
Gordon Parikh, SRF Consulting

Oscillations in traffic, also known as shockwaves, usually lead to reduced efficiency of traffic systems and introduce dangerous traffic conditions that are very likely to result in crashes. On freeway systems, such disturbances can produce crash-prone traffic conditions (CPCs). An analysis of traffic patterns during CPCs was performed and the signature of CPCs was traced through time and space from actual crashes. This study utilizes a multi-metric approach that creates a high-dimensional representation for traffic conditions with a high level of detail. A continuous measurement of crash probability is estimated based on this high dimensional representation. The crash probability estimate is incorporated in a real-time queue warning algorithm to provide consistent decisions of what, if any, warning message to communicate to the drivers. In collaboration with the Minnesota DOT, the developed Minnesota Queue Warning system (MN-QWARN), was integrated with the IRIS traffic management system, and implemented at a high crash frequency section of I-94 in Minneapolis. This paper summarizes the MN-QWARN algorithm and presents results from the first two years of the system’s operation. Based on extensive collection and analysis of ground truth data, it is shown that following the implementation of MN-QWARN, the freeway experienced approximately a 49% reduction of vehicle collisions and an 82% reduction in near-crash events. It should be noted that there was a 12% reduction in demand between the before and after periods so some of this benefit is also attributed to a reduction in congestion.

Zihe Zhang, University of Alabama
Jun Liu, University of Alabama
Xiaobing Li, University of Alabama
Asad J. Khattak, University of Tennessee

Incident duration models are often developed to assist the incident management and traveler information dissemination. With recent advances in data collection and management, enormous achieved incident data are now available for incident model development. However, a large volume of data may present challenges to practitioners, such as data processing and computation. Besides, data that span multiple years may have inconsistency issues due to the data collection environments and procedures. A practical question may arise in the incident modeling community – Do we really need that much data (“all-in”) to build models? If not, then how many data do we need? To answer these questions, this study aims to investigate the relationship between the data sample sizes and the reliability of incident duration analysis models. This study proposed and demonstrated a sample size determination framework through a case study using data of over 47,000 incidents. This study estimated handfuls of hazard-based duration models with varying sample sizes. The relationships between sample size and model performance along with estimate outcomes (i.e., coefficients and significance levels) were examined and visualized. The results showed that the variation of coefficients decreases as the sample size increases and becomes stabilized when the sample size reaches a critical threshold value. This critical threshold value may be the recommended sample size. The case study suggested a sample size of 6,500 to be enough for a reliable incident duration model. The critical value may vary significantly with different data and model specifications. More implications are discussed in the paper.

Paulina Reina, California State University, Fullerton

Empirical macroscopic analysis of traffic around freeway incidents revealed that rubbernecking on freeways is considerably prevalent and that it significantly affects freeway traffic. Particularly, a case study analysis revealed that queues instigated by rubbernecking presented significant queue lengths, congestion durations, and traffic delays. In addition, statistical models were estimated to identify exogenous traffic and incident characteristics associated with rubbernecking. It was observed that the presence of nearby on-ramps, high-occupancy vehicle lanes, and trucks were significantly correlated with the emergence of rubbernecking queues. Furthermore, an evaluation of rubbernecking models revealed key limitations of some statistical models in effectively identifying locations prone to rubbernecking.

Xiaohua Zhao, Beijing University of Technology
Yunjie Ju, Beijing University of Technology
haijian li (lihaijian@bjut.edu.cn), Beijing University of Technology
Jia Li, Beijing University of Technology
Jianming Ma, Texas Department of Transportation
Wenhui Dong, Beijing University of Technology

Drivers are required to make numerous rapid decisions at closely spaced interchange-tunnel section, which could result in drivers’ confusion, speed variation, and sudden lane change maneuvers. Therefore, closely spaced interchange-tunnel section is considered one of the most critical and challenging segments on freeways, but at which limited studies have investigated the factors affecting driving behavior. In this study, a driving simulator experiment was conducted to investigate risky driving behavior at interchange-tunnel section: recruiting a total of 39 participants, each of them performing eight different scenarios. Four driving behavior variables (i.e., average speed, speed SD, lateral position SD and acceleration SD) were extracred from the experimental data to investigate risky driving behavior with various paths, signs (i.e., interchange guide signs and tunnel distance signs), and arrow pavement marker. Besides, the research investigated the effects of drivers’ individual characteristics on driving behavior. The development of a series of liner mixed models with random effects reveals the influence factors for risk drivng behavior at closely spaced interchange-tunnel section. The results indicate that drivers who drive along the mainline at interchange-tunnel sections possess safer driving behavior compared with those who drive through the mainline and headed to the off-ramp. Removement of the third interchange guide sign, and adding the arrow pavement marker were recommended in this study. Providing tunnel distance signs in tunnel has a significant effect on risky driving behavior reduction. Finally, excessive signs and arrow pavement markers have a negative impact on driving behavior.

Ishtiak Ahmed (iahmed2@ncsu.edu), Institute for Transportation Research and Education (ITRE)
Alan Karr, AFK Analytics, LLC
Nagui Rouphail, North Carolina State University
Richard Chase, Institute for Transportation Research and Education
Shams Tanvir, California Polytechnic State University, San Luis Obispo

Abstract This study characterizes lane changing behavior under different congestion levels and identifies extreme lane changing behavior. Next Generation SIMulation (NGSIM) trajectories on US-101 in Los Angeles were used. Data were pre-processed to remove potential false-positive lane changes. Lane change frequency exhibited a reciprocal relationship with congestion levels, dropping by 33% as traffic speed dropped from 47 to 32 ft/s. Overall, average speed increased by 5.4 ft/s after changing lanes. However, this speed gain significantly diminished as congestion worsened. Further, the average speed of lane changing vehicles was 3.9 ft/s higher than those that executed no lane changes. Two metrics identified extreme lane changing behavior: critical time-to-line-crossing and lane changes per distance traveled. The lowest 1% critical times varied from 0.71–1.57 seconds. The highest 1% of lane change rates for all lane changing vehicles was found to be 2.5 lane changes per 1,000 ft. traveled. No drivers exhibited extreme behavior on both metrics.   Keywords: Driver behavior, extreme lane change maneuver, NGSIM, congestion level, lane change frequency, time-to-line-crossing.

Nishu Choudhary, Georgia Institute of Technology (Georgia Tech)
Angshuman Guin, Georgia Institute of Technology (Georgia Tech)
Michael Hunter, Georgia Institute of Technology (Georgia Tech)

Spot speed and vehicle count measurement has been the most widely accepted performance monitoring method for traffic operations data collection by transportation agencies. Spot-speed based and cumulative-count based delay estimation methods are typically deployed by practitioners and researchers alike for rapid estimation of delays as a precursor to congestion mitigation. In this paper, these commonly used incident-induced delay estimation methodologies, that are based on queuing theory or shockwave analysis models, are reviewed and validated against microscopic simulation of a real-life incident. For the simulation model, traffic data was obtained through the local Traffic Management Center’s detection system and the incident timeline was constructed using incident logs. The comparison revealed challenges related to noisy data and the failure of spot-speed measurements to adequately capture heterogeneity in congested traffic, which rendered the methodologies impractical for field use. In the absence of any alternative method to accurately quantify delay within the constraints of field observational data, a regression model was developed using data from a non-exhaustive set of incident scenarios simulated using Vissim®, to help obtain rapid estimates of delays for incidents with varying characteristics occurring under varying base conditions. This regression model can aid in resource allocation for efficient incident management and identification of influence factors.

Henrick Haule, Florida International University
Priyanka Alluri, Florida International University
Thobias Sando, University of North Florida

A traffic incident timeline constitutes seven intervals: detection, verification, dispatch, response, clearance, departure, and recovery time. Clearance time starts when responders arrive at the incident scene and ends when blocked lanes are open to traffic or when the roadway is cleared, whichever is earlier. Many studies have investigated the effect of spatial, temporal, and incident attributes on the clearance times. However, the impact of the incident timeline intervals that precede the clearance time has barely been studied. This study investigated the influence of the incident timeline intervals and other attributes on the clearance time. The incident data were collected for the years 2014 to 2017 on a freeway network in Duval County, Florida. A threshold regression (TR) model was applied to reveal the impact of factors on the clearance time and the unobserved condition at the incident scene when the first responder arrived. Also, the TR model results showed its advantages over the Cox proportional hazards (PH) model by including variables that were excluded in the Cox model because of violating the PH assumption. Results indicated that the incident scene condition when the first responder arrived was significantly affected by the verification and dispatch time, time of day, detection method, number of responding agencies, and severity. The TR model also indicated that the incident type, incident severity, verification, dispatch, and response time significantly influenced the clearance time. Transportation agencies could use the results to improve strategies and adopt policies that focus on reducing the length of the incident timeline intervals.

Eren Yuksel, University of South Florida
Robert Bertini, Oregon State University
Brian Staes, University of South Florida
Nikhil Menon, University of South Florida

There are many essential measured fundamental traffic data parameters including vehicle count, occupancy, and speed that can be used for transportation planning, design, operations, and performance monitoring/management in real time and over longer periods of time. Federal, state, regional/local agencies, and the private sector invest substantial resources toward collecting traffic data. Fundamental traffic data can also help to identify bottleneck locations, provide information to travelers, track economic impacts of travel demand, reveal traffic pattern changes due to incidents and construction, and can be converted to key performance measures. A small portion of this overarching data collection effort includes vehicle classification stations that provide truck counts (and lengths) for freight planning, pavement design, operations and enforcement purposes. There is a distinctive data stream available for the freeway system in the Portland, Oregon region, supported by the PORTAL data hub. In addition to housing continuous 20-sec vehicle count, speed, and occupancy data for more than 1,000 sensors since 2004, this system uniquely provides volume bins at 4 length-based classifications. This study analyzes traffic conditions along an 18-mile section of southbound I-5 in Portland. In order to reveal bottleneck locations and their activation times, oblique curves of cumulative vehicle count, time-mean speed, truck volume, and total vehicle length were utilized. Both bottleneck locations and their activation times were found to be reproducible from day to day. The availability of a continuous stream of vehicle length data revealed unique, high resolution, spatiotemporal features of freeway traffic hitherto unavailable to the researcher or practitioner.

Xiaoling Luo, Chongqing Jiaotong University
Xiaobo Ma (xiaoboma@email.arizona.edu), University of Arizona
Matthew Munden, Arizona Department of Transportation
Yao-Jan Wu, University of Arizona
Yangsheng Jiang, Southwest Jiaotong University

Queue length information is a critical input for ramp metering management. Based on accurate and reliable queue length, the inflow rate can be optimized to maximize the benefit of ramp metering. This paper proposes a queue length estimation method for metered on-ramps. In the proposed method, multiple data sources including INRIX data, controller event-based data, and loop detector data are used. The proposed method is based on the Resilient backpropagation neural network (RPROP-NN) model. In addition, the proposed method is enhanced by two techniques. One technique is to implement the decision tree to determine whether the queue length is larger than zero or not. Also, the second technique is to check whether the queue length reaches the ramp queue capacity by using the loop occupancy rate data. Three ramps along the SR-51 freeway in Phoenix, Arizona are selected to evaluate the proposed method. The proposed method is compared with the Kalman filter (KF)-based method, proposed in previous research. The results show that the average improvements over the KF-based method are 46.82% and 63.08% for the estimated accuracy and reliability, respectively.

Helen Waleczek, Ruhr-Universitat, Bochum
Justin Geistefeldt, Ruhr-Universitat, Bochum

On freeways with high traffic demand, hard shoulder running (HSR) can be an effective traffic management measure to increase the capacity by providing an additional travel lane during peak hours. While the positive effects of HSR on traffic flow quality were documented in a number of studies, the implications of HSR on road safety are more ambiguous. The paper presents results of a study in which accident data for seven freeway sections with HSR on freeways in Germany were analyzed over a long period of 13 years. All investigated sections are equipped with variable speed limits. The evaluation of crash frequencies on the investigated freeway sections revealed a high safety level. By combining crash data and traffic data it is shown that crash occurrence depends on the prevailing traffic conditions, with congestion being the most critical traffic state in terms of safety. Hence, safety improvements upstream of HSR segments can be related to the improved traffic flow and the reduction of congestion. In conclusion, the results of the investigation provide evidence that the implementation of HSR can improve road safety if state-of-the-art traffic control technology is applied and congestion can be relieved.

Qinhua Jiang, University of California, Los Angeles
Dong Nian, University of Cincinnati
Yi Guo, University of Cincinnati
Jiaqi Ma, University of California, Los Angeles

Adopting weather-responsive management strategies (WRMS) that use road weather data from Integrating Mobile Observations (IMO) and connected vehicle (CV) technologies will enable State and local agencies to be proactive and manage the system before negative impacts occur. This paper presented a case study that applies analysis, modeling, and simulation (AMS) tools to investigate the effectiveness of connected vehicle (CV) based Weather Responsive Management Strategies (WRMS). The goal is to address safety concerns on a real-world freeway corridor, a segment of the I–80 Connected Vehicle Testbed in Wyoming, under adverse weather conditions. This study simulates, evaluates, and discusses two CV-based WRMS applications: Traveler Information Messages (TIM) and Snowplow Pre-Positioning. The study designs operational alternatives for WRMS using CV data and develops an AMS tool with a microscopic traffic simulator to understand the effectiveness of two WRMS under different scenarios, including various CV market penetration, weather conditions, and WRMS algorithm settings. The simulation results indicate that CV-based WRMS applications can improve traffic safety performance, as measured by inverse time-to-collision (iTTC) with the increase of CV market penetration rate. The effectiveness is most dramatic under severe weather conditions. The results of this project provide operational insights that State and local transportation agencies may use in future strategic planning and real-time operations for their CV-based WRMS programs.

Yu Han (yuhan@seu.edu.cn), Southeast University
Meng Wang, Technische Universiteit Delft
Ziang He, Southeast University
Pan Liu, Southeast University

Variable speed limits (VSLs) are a common traffic control measure to resolve freeway jam waves. State-of-the-art  model predictive control (MPC) approaches of VSLs are developed based on Eulerian Lighthill-Whitham and Richards (LWR) models, where the decision variables are flows between road segments. It is difficult to implement constraints on speeds that are necessary in typical real-world speed limit systems, because converting  flow to speed  results in nonlinear and non-convex optimization formulations. In this paper, we develop a new MPC of VSLs based on a discrete Lagrangian LWR model, in which the decision variables are average speeds of vehicle groups. This allows formulating speed constraints as control constraints rather than state constraints in the MPC problem. The optimization of vehicle groups speeds is formulated as a linear programming problem which can be solved efficiently. The presented MPC controller is tested in a microscopic simulation environment. Simulation results show that the presented MPC resolves freeway jam waves efficiently with reasonable safety constraints implemented.