Freeway network traffic management based on distributed reinforcement learning

A distributed machine learning approach in traffic flow control and dynamic route guidance is presented. The problem domain, a freeway network traffic flow integration control application considers multiple objectives of system, is formulated as a distributed reinforcement learning problem. The Gini coefficient is adopted in this study as an indicator of equity. The DRL approach was implemented via a multi-agent control architecture where the decision agent was assigned to each of the on-ramp or VMS. The reward of each agent is simultaneously updating a single shared policy. The control strategy's effect is demonstrated through its application to the simple freeway network. Analyses of simulation results using this approach show the equity of the system have a significant improvement over traditional control, especially for the case of traffic peak hour. Using the DRL approach, the Gini coefficient of the network has been reduced by 28.99% compared to traditional method.