2020

Ramin Niroumand, North Carolina State University
Mehrdad Tajalli, United Parcel Service (UPS)
Leila Hajibabai, North Carolina State University
Ali Hajbabaie, North Carolina State University

This study develops a novel mixed-integer non-linear program to control the movements of a mix of connected-automated vehicles (CAVs) and connected human-driven vehicles (CHVs) through signalized intersections. The main idea is to introduce a new “blue” phase during which, CHVs must follow the vehicle immediately in front of them. During this phase, the trajectory of CAVs is continuously optimized by a central controller so that they lead groups of CHVs through an intersection. The movement of CHVs is incorporated in the optimization framework utilizing a customized linear car-following model. The optimization program will determine the signal indication of each lane-group in each time step. We have developed a receding horizon control framework to solve the problem. Case study results indicate that the proposed methodology successfully controls the mixed CAV-CHV traffic with various CAV market penetration rates and different demand levels. The results indicate that higher CAV market penetration rate induces more frequent blue phase indication compared to green-red signals. The proposed program reduces the total delay by 19.6% - 96.2% compared to fully-actuated signal control optimized in Vistro.

Zongyuan Wu, University of Southampton
Ben Waterson, University of Southampton
Bani Anvari, University College London

Urban person delay and congestion have becoming an increasing important issues. Connected vehicle (CV) technologies offer opportunities for managing urban traffic efficiently to reduce vehicle delays. The adaptive signal controls in CV environments are vehicle based controls, ignoring the importance of reducing person delay and improving person mobility in urban areas. This paper proposes an innovative Adaptive Person Based Signal Control Algorithm (APBSCA) to minimize person delay at isolated urbans. APBSCA is able to explore flexible phase combinations and stage sequences to find optimal signal timing solutions in certain prediction horizon. The vehicle information including positions, speeds and occupancy levels are collected through CV technology as data sources. A three-level dynamic programming approach is adopted in APBSCA to update the predictive departure time of every vehicle surrounding junctions, which is affected by network environments and signal decisions. APBSCA figures out optimal signal timing parameters that yield highest person delay saving values indicators at isolated junction over the prediction period and implement the corresponding signal timings. The results indicate that APBSCA have better results in reducing average person delay in vehicle in terms of high occupancy vehicles. APBSCA offers significantly average person delay reduction up to 55%. The proposed APBSCA indicates that person based controls have potential benefits in reducing person delay to consistent the future urban goals of improving person mobility over vehicle based controls by better utilizing CV data incorporating occupancy levels.

Russell Graves, University of Tennessee, Knoxville
H. M. Abdul Aziz, Kansas State University
Subhadeep Chakraborty, University of Tennessee

This work presents and demonstrates an algorithm for enabling multi-agent coordination in a non-stationary environment through a round-robin exchange of information between neighboring agents. This method is tested against traditional centralized techniques for optimizing throughput in a small-scale simulated traffic intersection for comparisons of computational load. The reported algorithm is also examined alongside non-communicating Q-learning in a large-scale simulation demonstrating an increase in performance over traditional methods with respect to vehicle throughput. Increase in computational load for each additional state-action pair was observed to be roughly linear versus the exponential growth in solution time experienced by centralized techniques. The improvement in performance over Q-learning was observed to be 29.79% for a 15 minute period with stable demand levels. Several simulations were carried out over 30 minute periods with shifting demand levels. During these tests, the overall increase in performance was 41.37%. Longer, roughly three hour tests were performed in which the vehicle arrival rates shifted roughly halfway through, t = 1.4hr. The number of vehicles served by the presented method was 166% that of Q-learning until the shift in arrival rate. At this time, the performance increase dropped to 117% but improved near the end of the simulation. Overall, the presented algorithm was shown to have relatively short, feasible computation times and improve system-level performance as compared to Q-learning even when vehicle arrival rates shift.

Zhen Yang, University of Michigan
Yiheng Feng, Purdue University
Henry Liu, University of Michigan

Traffic signal control plays an important role in regulating the traffic flows at signalized intersections. With the connected and automated vehicle (CAV) technology, the speed of the CAVs can be planned in a fuel-efficient way with the information from the traffic signals. Both traffic control strategies and vehicle trajectories significantly impact the performance of the intersection operation. In this paper, an integrated control scheme is proposed to optimize the traffic signals and control the CAVs in a mixed traffic condition with regular vehicles (RVs), and connected vehicles (CVs). A state transition diagram is designed for different modes of operation of CAVs including eco-trajectory planning, cooperative adaptive cruise control (CACC) and other CAV maneuvers. At the intersection, a mixed-integer linear programming (MILP) problem is formulated to optimize the signal timing plan and arrival time of CAVs, with the consideration of CACC platoons. Simulation results from a real-world intersection show the mobility and fuel economy benefits from the integration of eco-trajectory planning and signal optimization. The delay reduction can be as high as 18.7% under high penetration rate of CAVs, compared to actuated signal control. Sensitivity analysis confirms that the performance of the proposed model improves with the increase of penetration rates, in terms of total delay, fuel consumption as well as emissions.

Roger Lloret-Batlle, NISCI (MIT-SCALE)
Zi-Hao Wang, DiDi Chuxing
Minjun Liu, Didi-Chuxing
Jianfeng Zheng, Didi Chuxing Inc.
Henry Liu, University of Michigan

/* Layout-provided Styles */ div.standard { margin-bottom: 2ex; } We present a fast and exact method for traffic signal coordination in networks with different cycle lengths. The bandwidth maximization problem is formulated as a MILP program which is easy to solve and scalable, not even requiring a commercial MILP solver. In contrast with most current MILP signal coordination formulations, our bandwidths are not imposed by constraints, but instead, included in the objective function as piece-wise linear envelopes dependent on relative offsets and platoon arrival times. Bandwidths are defined for every pair of junction, and longer bandwidths on corridors arise from the optimization itself. The method simultaneously finds optimal offsets and phase sequence. We introduce by the first time the usage of sharp cycle basis in the loop constraints, stated in terms of link offsets only. Sharp cycle bases are used in the related Multi-Period Periodic Event Scheduling Problem (m-PESP). Node offsets are later retrieved by solving a mixed-linear diophantine equation system. We use trajectory data to incorporate accurate progression speeds and weights for the bandwidth terms in the objective function. The method is experimentally tested in the real-world city of Qingyuan, China.

Ruochen Hao, Tongji University
Wanjing Ma, Tongji University
Ling Wang, Tongji University
Chunhui Yu, Tongji University
Xiaolong Ma, Qingdao Hisense Transtech Co.

The Signal Phase and Timing (SPaT) message is an important input for research and applications of connected vehicles (CVs) and would assist CVs in speed control and improving traffic efficiency at intersections. However, actuated signal controllers, which are an important component of traffic control system, are not able to directly give the SPaT information since the SPaT is influenced by both signal control logic and real-time traffic demand. This study elaborates an estimation method based on the idea that an actuated signal controller would provide similar signal timing for similar traffic states. Thus, the quantitative description of traffic states is important. The traffic flow at each approaching lane was compared to water flow. The state of water flow can be indicated by state parameters, e.g., speed or height, and its energy, which includes kinetic energy and potential energy. Similar to the water flow, this paper proposed an energy model for traffic flow, and it also added queue length as an additional state parameter. Based on that, the traffic state of intersections can be extract and descripted. Then, a pattern recognition algorithm was developed to identify the most similar historical states and also their corresponding SPaTs, whose average is the estimated SPaT at the current time. The result shows that the average error is 3.1 seconds

Qinzheng Wang, Leidos, Inc.
Xianfeng Yang, University of Maryland, College Park
Jia Hu, Tongji University
Yongjie Lin, South China University of Technology

In recent years, the connected vehicle (CV) technology has reached a level of maturity and can be beneficial for traffic controls at signalized intersections. Enriched information provided by CVs could be utilized to estimate arterial origin-destination (OD) flow patterns and so as to support more efficient control algorithms. Under the CV environment, this paper proposes a methodology to adaptively change the signal offsets to provide dynamic progression to multiple critical path-flows. The critical paths, determined by CV trajectories, refer to the routes that connect those OD pairs with high traffic volumes. Following the multi-path progression principle, a real-time optimization model is constructed to design signal coordination plan with a control objective of maximizing green bandwidth along the determined critical paths. To solve the optimization problem, a solution algorithm based on dynamic programming is then proposed. Simulation-based experimental tests are also conducted in Vissim to evaluate the efficiency and effectiveness of the proposed control system. The results reveal that the travel time of vehicles along those critical paths by the proposed control system is much lower than the one by the conventional coordinated actuated system. In addition, the proposed model can greatly reduce the average delay and number of stops for all network flows.

Xiao Liang, Pennsylvania State University
S. Ilgin Guler, Pennsylvania State University
Vikash Gayah, Pennsylvania State University, University Park

This paper proposes a connected vehicle-based traffic signal control scheme that seeks to improve both vehicle and pedestrian operations. Real-time information on vehicle speeds and locations are combined with knowledge of pedestrian arrivals to optimize signal timings that minimize a weighted average of vehicle and pedestrian delays. Such real-time pedestrian information might be available using existing sensors—such as pedestrian push buttons or infrared detectors—as well as in a connected environment. The algorithm implements a rolling-horizon optimization framework that optimizes signal phase sequences over some period but only implements the first phase in the optimized sequence. The results reveal that considering pedestrians in the optimization can improve both pedestrian and vehicle delays compared to when pedestrians are ignored. Within the proposed framework, average vehicle delay increases and average pedestrian delay decreases as more weight is assigned to pedestrian delay in the optimization. In general, the average person delay can be minimized when the relative weight between vehicle and pedestrian delay is consistent with the average occupancy rate of cars. However, a different weight may be chosen to prioritize pedestrian movement, if desired. These results are robust under varying demand levels and demand patterns. The effectiveness of the algorithm decreases as the information level of pedestrian arrivals decreases, and the algorithm becomes ineffective when information from fewer than 60% pedestrians is available. However, the detection of more than 60% of pedestrians can likely be achieved using existing technologies and thus would be available in a connected environment.

Milan Zlatkovic, University of Wyoming
Zorica Cvijovic, Trihydro Corporation
Aleksandar Stevanovic, University of Pittsburgh
Yu Song, University of Wyoming

Connected Vehicles (CV) are an emerging technology with a large potential to improve traffic operations and safety. They allow for large amounts of information to be shared among vehicles and infrastructure. The limited number of field tests have proven that they can be used for different adaptive traffic control programs. However, there are still many areas that need to be covered through research. This paper develops and tests CV-based Signal Control Priority (SCP) algorithms for transit and freight vehicles. The algorithms are using the latitude/longitude (lat/lon) coordinates of CV vehicles and intersections to establish communication, share information and request priority. The SCP strategies are implemented through controllers’ built-in features and logic processor, using Econolite ASC/3 as a representative traffic signal controller. The tests were performed in VISSIM microsimulation with ASC/3 Software-in-the-Loop (SIL) controller emulator. State Street in Salt Lake City, UT, is selected as a test-case corridor. Five simulation scenarios were developed, including future CV freight and transit priority and a BRT implementation with a customized 3-level conditional priority. The paper shows that the developed communication protocol can be successfully used for special signal operations, and can be easily implemented in the field since it uses universal technologies. The developed SCP algorithms are successful in reducing delays for target vehicles in excess of 6%, without significant impacts on other traffic. The abundance of information obtained from CV vehicles can be used to further enhance control algorithms and create adaptive SCP programs.

Zihan Zhang, Tongji University
Jia Hu, Tongji University
Xin Li, Dalian Maritime University
Haibo Kuang, Dalian Maritime University
Xu Zhao, Dalian Maritime University
Peng Jia, Dalian Maritime University

This research presents an eco-approach controller with overtake capability. It overcomes the shortcomings of conventional eco approach and is able to: i) overtake slowing moving vehicles for ecological purpose; ii) optimize the travel duration approaching an intersection; iii) guarantee both fuel saving and vehicle’s mobility; iv) consider stochasticity of surrounding traffic; v) functional under partially connected and automated environment. It takes advantage of connected vehicle technology by taking in real-time vehicle and infrastructure information as optimization input, including obstacle vehicle’s speed and position, signal timing and more. The problem is formulated as optimal control and is solved by GPOPS. The nonlinear bicycle model is adopted as the system dynamics, and linearized to reduce computational burden. The stochasticity of surrounding traffic is considered as a probability distribution which is transformed into a linear constraint. Quantitative evaluation is conduct to compare the proposed controller against human drivers and baseline optimal controller which only has longitudinal automation. Sensitivity analysis is conducted for the arrival time of the ego Cooperative and Automated Vehicle (CAV). The evaluation results demonstrate that the proposed controller improves the fuel efficiency by 17.20%-70.12%, and outperforms two baseline controllers by 3.13-26.67% fuel saving. In addition, the greatest improvement in mobility is up to approximately 170%.

Keshuang Tang, Tongji University
Jiarong Yao, Tongji University
Chaopeng Tan, Tongji University
Jian Sun, Tongji University

Solely using trajectory data to estimate traffic flow parameters such as traffic volume for signalized intersections has received increasing attentions in recent years. Currently, traffic flow models or probabilistic models are mostly used to achieve this goal, which usually demands a comparatively high penetration rate of sampled vehicle trajectories or relies on site-specific assumptions of vehicle arrival patterns. To overcome this drawback, this study proposes a data-driven method based on the Singular Value Thresholding (SVT) algorithm for the cycle-based traffic volume estimation at signalized intersections. Unlike the existing methods, the proposed method doesn’t require site-specific assumptions of vehicle arrival distributions. Firstly, the trajectories captured in a cycle are used to calculate the arrival rate between the inter-arrival timepoints of consecutive trajectories. Then, an incomplete arrival rate matrix can be obtained after the arrival rate calculation of all the cycles during the study period and a rasterization standardization step to ensure equal sizes of arrival rate vector of each cycle. Thus, the volume estimation problem is transformed to be a matrix completion problem and is solved through the SVT algorithm. Evaluation is done by comparing with two existing methods through an empirical study. Results show that the Mean Absolute Percentile Error (MAPE) of the proposed method could reach 5.7% for the hourly volume estimation and 17.6% for the cycle-based volume estimation, given a penetration rate of 8.4%. In addition, the proposed method has better robustness than the two existing methods even under a penetration rate of less than 5%.

Minjun Liu, Didi Chuxing LLC
Jianfeng Zheng, Didi Chuxing LLC

This work proposes two bandwidth optimization models for arterial coordination. In contrast to existing models which mostly required a common cycle length, these two models are able to account for coordination with different cycle lengths at different intersections on an arterial, and thus would be more flexible for coordination design. As a tradeoff, the formulation leads to a large MILP problem which is much more difficult to solve due to much larger solution space. To address this challenge, we propose a heuristic algorithm extending dynamic programming to solve problem. Due to the reduction of time complexity from an exponential degree to polynomial one, the optimization can be applied for large-scale network. The second model extends first one by considering trajectory arrival distribution. Both models have been tested in real world, showing promising results to improve arterial coordination.

Satyajit Mondal, Indian Institute of Technology, Varanasi
Ankit Gupta, Indian Institute of Technology, Varanasi

Headway of vehicles during platoon dispersion at signalized intersection is one of the critical microscopic traffic characteristics in traffic flow theory. Also, the distribution of discharge headway of vehicles has a significant impact on the traffic generation process in most of the microsimulation approaches. However, few studies have investigated the vehicle discharge headway for interrupted flow at signalized intersection under mixed traffic conditions. Present study considers 20 intersections from six cities for comprehensive analysis of discharge headway. A box-and-whiskers plot is generated for discharge headway to quantify its reasonable profile. The diagram describes that headway of vehicle decreases with the queue dispersion. Although, a stable headway can be observed after the fifth vehicle position of a queue, giving a saturation headway of 2.05 sec/veh. Six types of continuous distributions are tested to model the discharge headway distribution. A statistical investigation is also performed to verify the best-fitted model for each vehicle position in a queue. The ranking of a best-fitted distribution is done for each vehicle position as per the statistical significance. This study demonstrates the discharge headway characteristics and distribution at each vehicle position, which can be useful for traffic flow analysis and especially for improving microsimulation models.

Shihong Huang, University of Michigan, Ann Arbor
Wai Wong, University of Michigan, Ann Arbor
Qi Chen, University of California, Irvine
Z. Mao, University of Michigan
Henry Liu, University of Michigan
Yiheng Feng, Purdue University

The rapid development of connected vehicle (CV) technology enables connectivity and data exchange between vehicles and transportation infrastructure. CV technology has great potentials in improving current traffic signal control systems with trajectory-based real-time signal control capabilities. However, the connectivity might also open a new door for cyber attacks, if the cyber-vulnerabilities of the control system can be identified and utilized by attackers. Unfortunately, most of the existing CV-based traffic signal control systems (CVTSCSs) are designed without considering potential cyber attacks. In order to effectively defend CVTSCSs against cyber attacks, cyber-vulnerability analysis is the prerequisite step. In this paper, we systematically analyze the structure of CVTSCSs and propose a generic analysis approach for identifying the cyber-vulnerabilities of CVTSCSs. The proposed approach comprises three major components: a list of traffic features, a surrogate model representing the control logic and a feature selection method identifying the critical features of a CVTSCS. The proposed analysis approach is tested with a selected CVTSCS that has been deployed at USDOT CV Pilot Development Programs. Two critical features are identified and validated by attack experiments, in which an attacker sends tampered trajectory data to manipulate the values of these two features and subsequently influence the signal control decisions. The risk of cyber attacks is realistically assessed by measuring delay increase. The experiment results demonstrate that the proposed approach is promising in identifying the cyber-vulnerabilities of CVTSCSs and providing a direction for developing defense strategies.

Jonathan Waddell, Wayne State University
Stephen Remias, INRIX
Jenna Kirsch, Wayne State University
Mohsen Kamyab, WSP

Probe vehicle trajectory data has the potential to transform the current practice of traffic signal optimization. Current scalable trajectory data is limited in both the penetration rate and the ping frequency, or length of time between vehicle waypoints. This paper introduces a methodology to create binary vehicle trajectories which can be used in a neural network to predict when vehicles will arrive at a virtual detector. The methodology allows for vehicles with ping frequencies of up to 60 seconds to be utilized for the optimization of offsets at signalized intersections. A nine signal corridor in west Michigan was used to test the proposed methodology. The neural network was compared to traditional linear interpolation strategies and found to improve the root mean squared error of the arrival times by up to 6.18 seconds. Using the virtual detector data stacked over time to optimize the offsets of the corridor resulted in 77% of the benefit of an offset optimization performed with continuously collected high resolution signal controller data. In the era of big data, this alternative approach can assist with the large-scale implementation of traffic signal performance measures for improved operations.