Model Predictive Control for Close Following Operation of High‐Speed Train Groups Driven by Personalized Long‐Term Trajectory Prediction

ABSTRACT

Grouping train control for high‐speed railways is an advanced control scheme. It allows the following trains to maintain a close headway based on the real‐time and accurate trajectory of the preceding train. However, due to nonlinear and uncertain dynamics as well as complex operating conditions, the trajectory of the preceding train is difficult to directly obtain or accurately predict over a long horizon. This paper proposes a personalized trajectory prediction framework that integrates Gaussian mixture model GMM‐based driving style recognition and long short‐term memory (LSTM) neural networks. It accounts for the individual driving characteristics of the preceding train, enabling accurate long‐term trajectory prediction under nonlinear dynamics. For the following train, a model predictive control (MPC)‐based close‐tracking scheme is designed that incorporates the predicted trajectory to extend the prediction horizon of the controller, while a safety margin is introduced to tighten safety constraints. Experimental results show that our approach maintains a 15‐s mean relative error (MRE) below 3.5%, achieving improvements of 32.9% and 43.3% over standard LSTM and Kalman filter (KF) approaches, respectively. Compared with a conventional MPC, our approach substantially reduces the required following distance while maintaining control stability and operational safety, shortening the tracking interval to 1520 m, particularly at high‐speed operating levels.