关键词: 改进深度强化学习, 电网侧储能, 奖励函数, 优化调度, 全生命周期

Abstract:

With the large-scale integration of new energy into the power grid, traditional scheduling modes struggle to address the high randomness and complexity of the system. The optimal scheduling of grid-side energy storage systems has become a key factor in enhancing the flexibility and reliability of the power grid. This paper proposes a peak-shaving control strategy for grid-side energy storage based on improved deep reinforcement learning: a multi-source data input layer is constructed by integrating renewable energy output, load demand, and energy storage equipment parameters, and a reward function that balances short-term peak-shaving benefits and long-term full-life cycle costs is designed, enabling the agent to learn the optimal scheduling strategy through interaction with the microgrid environment. Case studies based on a park-level microgrid test system show that, compared with traditional scheduling methods, this strategy reduces the full-life cycle cost by 11.9%-34.6%, extends the battery life by 22.55%-37.36%, increases the comprehensive renewable energy consumption rate to 92.3%, and reduces the peak-valley difference of the microgrid by 36.36%. This strategy provides a data-driven solution for the dynamic and intelligent management of grid-side energy storage systems in modern power grids, helping to improve the operational efficiency of the power grid and the ability to absorb renewable energy.

Key words: improved deep reinforcement learning, grid-side energy storage, reward function, optimized scheduling, full life cycle