Learning-Based Fault-Tolerant Traffic Management Algorithms for Intelligent Transportation Systems

点击翻译按钮获取中文摘要

This grant will support research on fault-tolerant intelligent transportation systems. Intelligent transportation systems have a strong potential for congestion mitigation and energy savings. However, such systems rely on a large number of sensors and communications devices deployed in complex environments. Therefore, these systems are prone to sensing faults (misreported, corrupted, or missing measurements), which can significantly compromise system efficiency. To address this challenge, this project will develop traffic management algorithms that detect and resolve sensing faults. Based on these algorithms, a paradigm of real-time, online fault-tolerant traffic management will be studied. This project will advance the knowledge about resiliency of intelligent transportation systems and provide ready-to-implement solutions. The solutions have the potential for mitigating the economic and environmental losses caused by sensing faults and improving cyber resilience of the traffic system. The expected results will provide hints on fault-tolerant design of other civil infrastructure systems as well. Graduate and undergraduate students from underrepresented groups will lead the research activities. Transportation agencies will be involved in discussion on concerns for actual deployment.The research activities jointly consider fault detection and resilient control for intelligent transportation systems. The modeling will synthesize physical law-based and learning-based approaches. Consequently, the models can integrate real-time information and adjust themselves to non-stationary environments with minimal human intervention. The objective will be approached via two research and one validation tasks. First, online fault detection and correction algorithms will be developed. These algorithms check consistency between observations from correlated sensors and between the observed data and physical law-based models, which contain prior empirical knowledge of the spatio-temporal correlation of sensor observations. Second, learning-based approaches will be incorporated into classical traffic control methods to design adaptive traffic management algorithms. The approach will provide guaranteed resiliency against sensing faults. Finally, the algorithms will be validated in a realistic micro-simulation testbed of a road network in New York City under a variety of sensing fault scenarios.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.