Multiagent-Game-Based Reinforcement Learning Energy Management Strategy for Flexible Traction Power Supply System With Energy Storage System

The conventional traction power supply system (TPSS) is limited in its ability to transport energy across regions due to the presence of section posts. In contrast, flexible TPSSs enable system-wide utilization of energy. However, electric locomotives face complex working conditions and experience drastic power fluctuations, making it crucial to address the efficient utilization of energy from multiple traction substations (SSs). This article proposes a multiagent-game-based reinforcement learning (MAG-RL) energy management strategy to facilitate collaborative energy interaction among neighboring SSs. Specifically, a Markov decision process for the energy management process is first established. On this basis, a distributed RL training framework is constructed to reduce the dimensionality of the state space. The multiagent energy game model is also constructed by analyzing the operation mechanism under different operation modes. Additionally, a sequential negotiation method is presented to quickly solve the energy game model. Real-time simulation platform testing indicates that the proposed method reduces the number of convergence iterations by 28.2% compared to the traditional <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Q</i>-learning approach. Compared to independent operation, the utilization efficiency of regenerative braking energy improves by 30.04%, reaching 93.28%, demonstrating a technical advantage over other strategies.