In order to improve the operation efficiency of urban rail transit and reduce energy consumption, a multi-objective train energy saving optimization model was established based on the train motion equation, parking accuracy and punctuality. On the basis of making full use of the stability of TD3 (Twin Delayed Deep Deterministic policy gradient) algorithm in processing the high-dimensional continuous action space and better adaptability in the face of complex environment, the maximum speed curve of the train is introduced. A reasonable action range is set to improve the efficiency of the algorithm and reduce the difficulty of solving the model. According to the train schedule, different single interval can be set up to allocate the running time and appropriate reward function within the allowable range of the total running time. The deep reinforcement learning method is used to optimize the train running curve under different running time in each interval by the agent choosing the action freely within a certain range. The simulation experiment is designed on the background of 5 stations and 4 sections of Beijing Metro Line 2, and the improved DDPG algorithm and TD3 algorithm are compared. The efficiency of the improved algorithm is verified by setting up three different energy-saving train running curves under the single interval distributable running time. The results show that the energy consumption of train traction can be reduced to a certain extent by adjusting the allotted time of the interval. However, the energy saving benefits brought by selecting too large allot time are not significant, and will greatly affect the normal operation of trains. Therefore, adjusting the running time of the train in a single interval can effectively save the overall traction energy consumption of the train in multiple intervals, and the maximum energy saving can reach 18.07% compared with the traditional control method. The research results can provide a feasible control method for the energy-saving operation of urban rail trains in multiple sections.
