Objective There are challenges in urban rail transit trains, such as complex dynamic systems and higher difficulties in control system development. Therefore, it is necessary to propose a train speed control method based on a hierarchical architecture.

Method A single-particle model for train operation is built, which provides a foundation for the design and verification of the train speed control method. The upper-layer controller adopts the MPC (Model Predictive Control) algorithm, which can accurately solve the desired acceleration based on the reference speed and real-time feedback information of train operation status. Meanwhile, the MPC algorithm is improved in consideration of constraints such as train actuator delay and parking accuracy to enhance the train operational stability and parking accuracy. The lower-layer controller can rea-lize the tracking control of the train's desired acceleration by outputting the control levels. The specific calculation method is as follows: the train control levels are solved by designing a working condition switching strategy and an inverse dynamic equation, and the PID (proportional-integral-derivative) control algorithm is introduced to correct the acceleration tracking error. A simulation model of the train speed control method is built on the MATLAB/SIMULINK platform. The simulation results of normal operation are comparatively analyzed with those of the upper-layer controller using the traditional MPC algorithm, and a resistance disturbance simulation test is carried out for the proposed speed control method. Finally, the effectiveness of the proposed method is verified.

Result & Conclusion Based on the decoupling of the train operation process, the proposed method combines the improved MPC algorithm and PID algorithm to form a hierarchical control architecture, achieving precise speed control of the complex train system and enhancing the comfort level during train operation. Meanwhile, the improved MPC algorithm significantly improves the train's parking accuracy without excessively increasing the arrival error time.