An intelligent train control approach based on the monte carlo reinforcement learning algorithm

Traditional research on the train control problem use the determined train model and the control method cannot deal with the variable parameters in the complex environment. This paper proposes a new train control approach by analyzing the operation data and the reinforcement learning method instead of using a detailed train model. Specifically, a similarity-based sampling method is introduced to sample segments of train operation data, which is used to predict the running state of the train. With the operation data, the reward of each segment for the train operation can be evaluated. Then, the monte carlo reinforcement learning method is applied to select the optimal solution for the train control decision. Finally, a case study is conducted to illustrate the effectiveness of the proposed approach based on the practical operation data of Yizhuang line.