A convolutional neural network driven suspension control strategy to enhance sustainability of high-speed trains

Abstract

With the accelerated expansion of rail networks and the increase in operation speeds, railway undertakings are under considerable pressure to curtail energy consumption of high-speed trains to achieve sustainability goals, while still maintaining passenger satisfaction. For addressing this challenge, a convolutional neural network driven control strategy is proposed for the suspension system of high-speed vehicles to simultaneously reduce energy consumption and carbody vibration on curved tracks. Firstly, a co-simulation platform is established between the multibody dynamics simulation software and MATLAB/Simulink, and a series of running conditions are designed. Based on the co-simulation results, the roles that train’s velocity, track curvature, and scale factor of the Skyhook controller play in energy efficiency and lateral carbody vibration are systematically studied. Subsequently, a convolutional neural network is constructed based on the simulation data to predict the energy consumption and riding comfort under complex operation scenarios. In conjunction with the neural network algorithm, a bi-objective optimization model is further developed and solved to adjust scale factor of the Skyhook controller according to different running conditions. The optimization results indicate that energy consumption and lateral vibration of a high-speed train on curved tracks can be respectively reduced by up to 15.90 % and 47.78 % through employment of the proposed control strategy.