Self-learning adaptive traffic signal control for real-time safety optimization

Abstract

Adaptive traffic signal control (ATSC) is a promising technique to improve the efficiency of signalized intersections, especially in the era of connected vehicles (CVs) when real-time information on vehicle positions and trajectories is available. Numerous ATSC algorithms have been proposed to accommodate real-time traffic conditions and optimize traffic efficiency. The common objective of these algorithms is to minimize total delay, decrease queue length, or maximize vehicle throughput. Despite their positive impacts on traffic mobility, the existing ATSC algorithms do not consider optimizing traffic safety. This is most likely due to the lack of tools to evaluate safety in real time. However, recent research has developed various real-time safety models for signalized intersections. These models can be used to evaluate safety in real time using dynamic traffic parameters, such as traffic volume, shock wave characteristics, and platoon ratio. Evaluating safety in real time can enable developing ATSC strategies for real-time safety optimization. In this paper, we present a novel self-learning ATSC algorithm to optimize the safety of signalized intersections. The algorithm was developed using the Reinforcement Learning (RL) approach and was trained using the simulation platform VISSIM. The trained algorithm was then validated using real-world traffic data obtained from two signalized intersections in the city of Surrey, British Columbia. Compared to the traditional actuated signal control system, the proposed algorithm reduces traffic conflicts by approximately 40 %. Moreover, the proposed ATSC algorithm was tested under various market penetration rates (MPRs) of CVs. The results showed that 90 % and 50 % of the algorithm’s safety benefits can be achieved at MPR values of 50 % and 30 %, respectively. To the best of the authors’ knowledge, this is the first self-learning ATSC algorithm that optimizes traffic safety in real time.