Research on Charge and Discharge Strategies Optimization based on Reinforcement Learning

At present, it is anticipated that electric vehicles suit to address the dual challenges of energy shortage and environmental pollution, due to the utilisation of clean energy, high efficiency, and exemplary energy storage characteristics. Nevertheless, the extensive and unstructured charging of electric vehicles is likely to intensify the peak-to-valley and overall variance of the daily load curve. Although the conventional time-of-use pricing approach has served to alleviate this issue to a degree, it has simultaneously precipitated the emergence of new load peaks. In this work, we propose an optimisation method for charging and discharging strategies for electric vehicles based on reinforcement learning. A model has been developed which employs the Proximal Policy Optimization (PPO) algorithm within the deep reinforcement learning framework to identify the optimal charging and discharging strategy through interaction with the grid environment. In contrast to conventional methodologies, we have incorporated a self-attention mechanism into the PPO algorithm to enhance the efficacy with which the timing characteristics of electricity prices and grid loads can be captured. The model considers a range of factors, including electricity price, grid load, vehicle usage mode and user preference, and employs a multi-objective reward function to achieve a balance between charging cost, user comfort and grid stability. By monitoring the status of the grid in real time and anticipating future developments, the model is able to make intelligent decisions regarding charging and discharging, thereby achieving the dual objectives of peak shaving and valley filling while reducing user costs. The results of the experimental analysis demonstrate that our method is markedly more effective than the traditional optimization strategy in reducing charging costs and alleviating peak loads, and has considerable practical application value.