Deep Learning-Based Model Predictive Control for Virtual Coupling Railways Operation
Haoxiang Su, Ming Chai, Lijie Chen, Jidong Lv
Abstract
This paper presents an approach in train control systems for virtual coupling. A deep learning-based model predictive control is proposed for operations of following trains in virtual coupling. The proposed MPC control method includes a long and short-term memory (LSTM) neural network and an optimal controller. The LSTM neural network is used to predict control objectives of the preceding train's operation status. The MPC-based controller with the prediction results for the following train is presented. This paper compares the proposed approach with conventional MPC by using a concrete example of a metro line. The experimental results show that with a deep learning-based prediction of preceding trains, the following train has better tracking accuracy to improve the stability of train convoys.