In the face of fossil fuel depletion, improving energy efficiency is crucial to sustainable development. In AC electrified railway systems, the efficient allocation of Regenerative Braking Energy (RBE) is important to improve the energy efficiency and realize low-carbon transportation. Its level of optimization directly affects the system’s energy recovery efficiency and economic viability, making it highly significant for the development of sustainable transportation. However, current research in this field remains relatively limited, so it requires further exploration of more intelligent energy allocation strategies. To improve the utilization efficiency of regenerative braking energy in AC electrified railway systems, this paper proposes a power allocation method for substations based on Q-learning. The method first involves discretized modeling of the substation's energy state. At the same time, this paper has designed the action space according to the operational characteristics of the power distribution unit and introduced an ε-greedy strategy combined with an energy-saving-oriented reward function. All the above can guide the agent to progressively learn the optimal energy allocation policy through interaction. Simulation verification was conducted using a typical three-substation system. The results show that the Q-table effectively converges. And the learned policy exhibits good energy utilization efficiency and system adaptability in dynamic environments.  The overall utilization rate of regenerative braking energy reaches 60%, verifying the feasibility and effectiveness of the proposed method in improving energy recovery efficiency

Reinforcement learning, power allocation, substations, regenerative braking energy