An intelligent train operation algorithm via gradient descent method and driver's experience

Abstract

Most existing train control methods aim to track the target velocity curve offline, which may cause the frequent shit of the controller output, reduced comfort of passengers and increased energy consumption etc. Different from the previous control strategies, this paper presents a new algorithm without using the model information and the offline target velocity curve. The new algorithm is a data-driven intelligent train operation (ITO) algorithm which uses driver's experience to obtain the control strategy and employs input-output data to online optimize by gradient descent method. The proposed algorithm is tested in a Matlab/Simulink simulation model using the actual data from Beijing subway Yizhuang line. Compared with Proportion-integral-derivative(PID), this algorithm is better with less energy consumption, higher comfort, and parking precision and it meets the dynamic adjustment of running time. Moreover, the results of the ITO algorithm look like driver's situation both on trajectory and operation mode conversion.