Traffic Congestion Prediction and Intelligent Signalling Based on Markov Decision Process and Reinforcement Learning

Abstract

As the advancement of vehicular traffic, traffic congestion alleviation is desperately required in urban cities. A fixed duration traffic light in an intersection can make a few empty lanes and at the same time create other congested lanes. Dynamic scheduling of the signals is renowned as a solution for traffic congestion mitigation in urban areas. Static phase timing is not an optimal solution for reducing the congestion at the signals. So there is a pressing need of efficient algorithms for congestion prediction by considering historical and real time traffic data. The proposed work adopts to optimize a standard traffic a junction of two roads, one with North–South orientation and other with East–West orientation stop light dynamically with reinforcement learning and with markov decision process. It considers inflow and out flow of traffic at each lines and also waiting time of the vehicle for scheduling the signal timings. By using this method the overall waiting time of vehicles considerably reduced.