[Schedule] TRBAM-2022 Presentation and Poster

January 10, Mon

Here’re the general information for presentation and poster sessions of our group:

  • [Poster Presentation] B682 – Stochastic Calibration of Automated Vehicle Car-Following Control: An Approximate Bayesian Computation Approach – TRBAM-22-01658

Ghazaleh Jafarsalehi (University of California, Davis), Yang Zhou, Soyoung Ahn, John D. Lee

[Mon 10:30 am, January 10] 1093 – Traffic Flow Theory, Part 1: Network Modeling and Control (Part 2, Session 1178; Part 3, Session 1337; Part 4, Session 1371)

This paper presents a stochastic calibration method based on Approximate Bayesian Computation (ABC) to calibrate the parameters of two well-known adaptive cruise control (ACC) models: the linear controller and Model Preditive Controller (MPC)-based controller. The method is likelihood-free, where the likelihood function is replaced by simulation of the model followed by a comparison with the observed data to approximate the posterior distribution. This structure affords flexibility to calibrate posterior joint distributions of complex models, even those without analytical closed forms such as MPC. The proposed calibration method was able to generate meaningful posterior distributions: the posterior marginal distributions for some key parameters displayed an obvious cluster with a significant peak, and the posterior joint distributions revealed correlations among some key parameters. Validation results suggest that more than 80% of simulated trajectories successfully reproduced the empirical data within an acceptable error. Nonetheless, some systematic errors were noticeable in the simulated vs. observed vehicle position, potentially attributed to data limitation (e.g., limited coverage and highly dynamic traffic conditions) and possibly model mismatch.

 

January 11, Tue

  • [Presentation] Tribal Crash Reporting System Improvements in Wisconsin – TRBAM-22-02539

Tianyi Chen, Haotian Shi, Glenn Vorhes, Steven T. Parker, David A. Noyce

[Tue 10:30 am, January 11] 1257 – Tribal Crash Reporting

Safety programs for analyzing and reducing crashes are important in tribal lands due to high crash rates and crash severities. The comprehensive tribal crash reporting and tribal crash data quality are imperative to apply successfully for funds and develop effective tribal safety programs. With eleven tribal nations located within the State of Wisconsin, there is a strong and long-standing interest in analyzing Wisconsin tribal crashes and improving traffic safety within Wisconsin tribal lands. The first step to conduct the tribal safety analysis is to systematically validate the quality of tribal crash data in the current Wisconsin crash database system. This paper investigates the advantages and deficiencies of the current Wisconsin crash system within respect to tribal crashes and proposes several recommendations for the crash report improvement. Specifically, we verify the effectiveness and quality of Wisconsin crash datasets provided by the Traffic Operations and Safety (TOPS) Laboratory at the University of Wisconsin-Madison. Quantitative and spatial data analyses are presented using the Geographic Information Systems to validate the rationality of updated elements and the data quality of the tribal crashes in terms of three crash attributes: location type, jurisdiction, and law enforcement agency. The results indicate that these tribal attributes are coupled which has a negative effect on data quality for further safety analysis. The findings recommend that adding an independent new subfield in crash classification for tribal element and adding a specific tribal road type in roadway data elements would be more effective and reasonable for tribal crash reporting and analysis.

  • [Poster Presentation] B631 – A Distributed Deep Reinforcement Learning Based Integrated Dynamic Bus Control System in a Connected Environment – TRBAM-22-00882

Haotian Shi, Qinghui Nie, Sicheng Fu, Xin Wang, Yang Zhou, Bin Ran

[Tue 1:30 pm, January 11] 1313 – Bus Transit Operations, Intelligent Transportation Systems, and Technology

The bus bunching problem caused by the uncertain inter-station travel time and passenger demand rate is a critical issue that impairs transit efficiency. Most current bus control studies focus on single or combined strategies while ignoring the bus system’s real-time environmental information. This paper proposed a distributed deep reinforcement learning (DRL) based generic bus dynamic control method to solve the bus bunching problem by maintaining the schedule adherence, headway regularity, and achieving the consensus in the multi-agent system. This study built a bus system that utilizes the bus historical and traffic information by incorporating these characteristics into the environment. After that, a distributed DRL-based bus dynamic control strategy is developed based on the bus system, enabling each bus to adjust its motion by any generic method utilizing the weighted downstream buses’ information. Regarding the training process, a distributed proximal policy optimization algorithm is adopted for improving the converging performance. Simulated experiments are conducted to verify the control performance, robustness, feasibility, resilience, and generalization, which shows our strategy can significantly reduce the schedule and headway deviations, prevent the accumulation of deviation downstream, and avoid bus bunching.

  • [Poster Presentation] B654 – Work Zone Crash Occurrence Prediction Based on Planning Stage Work Zone Configurations Using an Artificial Neural Network – TRBAM-22-03731

Yang Cheng, Keshu Wu, Hanchu Li (Southeast University), Steven T. Parker, Bin Ran, David A. Noyce

[Tue 4:00 pm, January 11] 1340 – Advancing New Methods and Data

Work zones are essential to maintain and improve the nations road infrastructure. However, work zones affect traffic safety, and crashes and fatalities associated with work zones in the U.S. have increased substantially. Most existing work zone crash studies are not able to support the improvement of work zone planning and configuration, despite providing insights about individual crash level attributes. This study proposes an artificial neural network (ANN) based approach to predict the crash occurrence in work zones only using work zone configurations and design parameters. The goal is to explore whether using simple work zone configuration features available at the planning stage as the input can achieve satisfying work zone crash prediction. The performance of the proposed model is satisfying and comparable with existing studies using more comprehensive features. The proposed approach, early at the work zone design and planning stage, can provide designers and decision-makers with quick work zone safety evaluation for design alternatives and suggest extra resources and attention needed.

 

January 12, Wed

  • [Presentation]Data-Driven Stochastic String Stability Analysis for Car-Following Control of Automated Vehicles – TRBAM-22-01359

Yang Zhou, Qian Chen (Southeast University), Soyoung Ahn, Jiwan Jiang, Ghazaleh Jafarsalehi (University of California, Davis)

[Wed 10:30 am, January 12] 1394 – Global Overview of Recent Advances in Traffic Flow Theory

This paper presents a data-driven, stochastic strong stability analysis method, considering measurement noise and estimation error. The method can be applied to complex or unknown controllers, for which a traditional theoretical string stability analysis is not feasible. Specifically, Welchs method is developed to extract the empirical transfer function from vehicle trajectories. Based on the empirical transfer function, we propose a stochastic strong string stability criterion to evaluate string stability of car-following control in a probabilistic fashion. The proposed method is validated for estimation consistency of transfer function norms and string stability classification via synthetic data-based simulation. The method is applied to field data of partially automated vehicles with adaptive cruise control (ACC) function on the market. Findings reveal that all tested vehicles are strong string unstable, particularly for low frequency disturbances (< 0.5 Hz). Beyond the stability classification, the method provides insight into string stability probability for each frequency, which is a unique contribution to the stability analysis of carfollowing control.