Perimeter Traffic Flow Control for a Multi-Region Large-Scale Traffic Network With Markov Decision Process

The coordination of traffic flow among regions is necessary for a large-scale road traffic network to avoid local congestions and improve the overall traffic efficiency. In this paper, by incorporating the random characteristic of traffic flow, we formulate the problem of perimeter traffic flow control for a multi-region traffic network as a Markov decision process with adaptive state definition. Based on stochastic macroscopic fundamental diagrams (MFD) of regions, a state transition probability model is proposed to describe the state changes of the multi-region traffic network under different perimeter control policies. With the stochastic MFD-based state transition probabilities rather than counting from the historical data, a policy iteration algorithm with perturbation analysis is introduced to get the optimal perimeter control policy in real-time without the requirement of online or offline learning. The proposed method is compared with the classic perimeter control methods by simulation, which indicates its effectiveness in mitigating the congestion and improving the network throughput, as well as the promising implementation prospect.