Multi-Scale Signal-Vehicle Coupled Control with Connected and Autonomous Vehicles

Abstract

With the rise of connected and autonomous vehicles (CAV) and vehicle-to-everything communication technologies, advanced urban traffic control (UTC) has been receiving increasing attention due to the abundant data and control freedom brought by the new techniques. The UTC is a multi-scale problem in both spatial and temporal domains. The goal of this study is to investigate the multi-scale UTC problems in the environment of CAVs. Firstly, we propose a general modeling and control framework for the multi-scale UTC problem and identify the key research needs. We then use this framework to explicitly model, analyze, and solve the two-scale signal-vehicle coupled control problem in the environment of single intersection with 100% CAV penetration, and demonstrate the performance of the proposed method using numerical experiments. Then, we extend the signal-vehicle coupled control algorithm to the environment of multiple intersections with mixed traffic flows of human driven vehicles and CAVs. Considering that for explicit model based methods, the computation burden increases with the size of the traffic network, we develop an imitation learning enhanced method to improve the algorithm efficiency. We conduct various numerical experiments to validate the performance of the proposed model-based and learning-enhanced algorithms.