Traffic Energy Saving Control Based on Reinforcement Learning

Abstract

Train energy-saving operation control is a research hotspot in the field of urban rail transit energy-saving. By strengthening the perception ability and decision-making ability of the learning algorithm, this paper puts forward a new idea for the train energy saving control in urban rail transit under the condition of ensuring safety, comfort, real-time and punctuality. To be specific, the following work is done in this paper: (1) Study the related knowledge of train dynamics, establish the train traction model and the train running resistance model and complete the force analysis of the train motion process; (2) Study the knowledge related to energy consumption of train operation and establish the calculation model of energy consumption of trains within the interval; (3) Study the knowledge related to reinforcement learning algorithm, transform the train operation control process into Markov decision process, establish the three elements of reinforcement learning algorithm, and solve the train energy saving control problem by programming. Through simulation, the method proposed in this paper can reduce energy consumption by 13%–17% under the constraints of safety and punctuality.