A Data-Driven Train Control Model for High-Speed Trains Based on LAG-LSTM

Abstract

In high-speed railways (HSRs), the construction of accurate train control models (TCMs) is the key to design train operation algorithms and real-time train rescheduling decisions. Due to the long mileage, frequent switching of working conditions and variation of external influencing factors (such as weather, temperature), traditional physical-driven models usually fail to reflect the “true” dynamics of high-speed trains (HSTs). Although some data-driven deep learning models have been proposed recently to realize environment adaptation, but they are all “black box” models, which cannot explain how the input of the model affects the output. In order to overcome above issues, LAG-LSTM model is constructed based on historical data collected at Beijing-Shanghai HSR. Specifically, considering the delay of control variables and the lack of state variables of HSTs, LAG-LSTM is divided into three modules, i.e. time-delay variable module (TDVM), state variable enhancement module (SVEM) and Pre-LSTM module. The simulation results based on field data show that LAG-LSTM evidently outperforms existing models and can accurately predict the trajectory of HSTs.