Automated Adaptive Traffic Corridor Control Using Reinforcement Learning

Abstract

Advancements in intelligent transportation systems and communication technology could considerably reduce delay and congestion through an array of networkwide traffic control and management strategies. The two most promising control tools for freeway corridors are traffic-responsive ramp metering and dynamic traffic diversion using variable message signs (VMSs). The use of these control methods independently could limit their usefulness. Therefore, integrated corridor control by using ramp metering and VMS diversion simultaneously could be beneficial. Administration of freeways and adjacent arterials often falls under different jurisdictional authorities. Lack of coordination among those authorities caused by lack of means for information exchange or “institutional gridlock” could hinder the full potential of technically possible integrated control. Fully automating corridor control could alleviate this problem. Research was conducted to develop a self-learning adaptive integrated freeway–arterial corridor control for both recurring and nonrecurring congestion. Reinforcement learning, an artificial intelligence method for machine learning, is used to provide a single, multiple, or integrated optimal control agent for a freeway or freeway–arterial corridor for both recurrent and nonrecurrent congestion. The microsimulation tool Paramics, which has been used to train and evaluate the agent in an offline mode within a simulated environment, is described. Results from various simulation case studies in the Toronto, Canada, area are encouraging and have demonstrated the effectiveness and superiority of the technique.