An Energy Management Strategy for Traction Substation with ESS Based on Reinforcement Learning

Abstract

The traction power supply system (TPSS) consumes a large amount of electrical energy for locomotive traction every year. To effectively utilize the regenerative braking energy of locomotives and reduce the overall traction energy consumption, the reinforcement learning (RL) method is introduced into energy management strategy in this paper. A reinforcement-learning framework for energy management is constructed to enhance energy utilization efficiency in traction scenario and regenerative braking scenario respectively, and to achieve the effect of adaptive charging and discharging by setting multiple reward functions. Meanwhile, due to the model-free energy management strategy, which avoids modeling the traditional circuit model and has the ability to update online, it is more flexible compared to the rule-based class approach. The simulation test results show that compared with the traditional rule-based energy management method, the RL-based energy management strategy can effectively improve the utilization efficiency in different electric locomotive operating scenarios, reduce the power impact of the traction power supply system, and improve the system operation economy.