Proposing an Optimal Control Method for Energy Consumption Optimization in Railway Signalling Systems Using Reinforcement Learning

Nowadays, the optimization of energy consumption in public transportation systems is a serious issue. Since a large part of energy in transportation systems is consumed by subways, a new approach has been proposed for optimal control of a train to reduce energy consumption. The proposed model is based on the Reinforcement Learning algorithm. It is assumed that a train moves between two stations along a line with non-constant gradient, curve, and speed limits. Moreover, the train should complete its journey within a given time interval. The Reinforcement Learning of States, Actions, and Rewards are based on the selected Actions. In the proposed method, the train States are the velocity and position of the train, and the Action is acceleration or coasting motion. Unlike the former techniques, most stages of optimization in this method are offline and implemented only once for any route. Following the formation of the reward matrix, we could use this method in an online form and then the speed profile could be produced at a minimum time. The simulations of the proposed method are implemented in MATLAB and finally compared with those of the Genetic Algorithm.