| 摘 要: 针对现有电氢混合储能系统调度研究多聚焦单一经济目标，难以兼顾电池寿命与储能利润等问题，本文构建了考虑经济性和电池寿命管理的多目标优化模型。为在多目标之间实现有效协调，本文提出了一种结合动作过滤策略的近端策略优化(proximal policy optimization, PPO)算法，该算法通过动作过滤抑制过度充放电行为，并基于多目标奖励函数实现电池退化率最小化与利润最大化的平衡。在山东地区数据集上的实验结果表明，相较于其他五种多目标深度强化学习算法，所提出方法在优化多个目标方面均表现出更好的性能，从而验证了所构建模型的可行性与所提算法的有效性。 |
| 关键词: 深度强化学习 电氢混合储能系统 电池退化 多目标优化 |
|
中图分类号:
文献标识码:
|
| 基金项目: 教育部人文社会科学研究青年基金项目（No.21YJC630087） |
|
| Research on Multi-objective Scheduling of Electro-hydrogen Hybrid Energy Storage System Based on Proximal Policy Optimization Algorithm |
|
Shao Qian
|
University of Shanghai for Science and Technology
|
| Abstract: To address the problems that existing studies on the dispatching of electric-hydrogen hybrid energy storage systems mostly focus on a single economic objective and find it difficult to balance battery lifespan against hydrogen storage costs, this paper constructs a multi-objective optimization model that takes into account economy as well as the management of battery lifespan. To achieve effective coordination among multiple objectives, this paper proposes a multi-objective Proximal Policy Optimization (PPO) algorithm integrated with an action filtering strategy. This algorithm suppresses excessive charging and discharging behaviors via action filtering, and realizes a balance between minimizing battery degradation rate and maximizing profit based on a multi-objective reward function. Experimental results on the dataset from Shandong region demonstrate that, compared with five other multi-objective deep reinforcement learning algorithms, the proposed method exhibits superior performance in optimizing multiple objectives, thus verifying the feasibility of the constructed model and the effectiveness of the proposed algorithm. |
| Keywords: Deep reinforcement learning hybrid electric-hydrogen energy storage system battery degradation multi-objective optimization |
