风储联合电站实时自调度的高效深度确定性策略梯度算法  

Efficient Deep Deterministic Policy Gradient Algorithm for Real-Time Self-Dispatch of Wind-Storage Power Plant

作　　者：宋煜浩 魏韡[1] 黄少伟[1] 吴启仁 梅生伟[1] Song Yuhao;Wei Wei;Huang Shaowei;Wu Qiren;Mei Shengwei(Department of Electrical Engineering,Tsinghua University,Beijing,100084,China;China Three Gorges Renewables(Group)Co.Ltd,Beijing,101100,China)

机构地区：[1]清华大学电机工程与应用电子系,北京100084 [2]中国三峡新能源(集团)股份有限公司,北京101100

出　　处：《电工技术学报》2022年第23期5987-5999,共13页Transactions of China Electrotechnical Society

基　　金：中国长江三峡集团有限公司科研项目资助(202003128)。

摘　　要：发展风电等可再生能源对于实现双碳目标具有重要意义,风储联合电站是未来风电接入电网的主要形式。该文研究发电侧商业化运行的风储联合电站的实时自调度问题,目标是使自身的期望收益最大化。由于场站级风电预测误差较大,独立发电商信息有限,难以准确预测电网电价,风储联合电站实时自调度面临多重不确定性,极具挑战。该文提出高效深度确定性策略梯度(DDPG)算法求取风储联合电站实时自调度策略,实现不依赖预测的场站级在线决策。首先通过Lyapunov优化构建基础策略,得到一个较好的但未必是局部最优的策略;然后,采用基础策略预生成样本,用于初始化经验库,提升搜索效率;接着,应用引入专家机制的DDPG算法,可以训练得到局部最优的自调度策略;最后,算例分析表明,相比于基础调度策略和经典DDPG,该文所提方法能有效提升风储联合电站的平均收益。The development of wind power and other renewable energy is of great significance to achieve the dual carbon goal,and the wind-storage power plant is the main form of wind power connected to the power grid in the future.This paper studies the real-time self-dispatch problem of the wind-storage power plant commercialized on the generating side,with the goal of maximizing its expected income.Due to the large prediction error of the field-level wind power and the difficulty in accurately predicting the electricity price of the grid due to the limited information of independent power producers,the realtime self-dispatch of the wind-storage power plant is faced with multiple uncertainties,which is extremely challenging.In this paper,an efficient DDPG algorithm was proposed to solve the real-time self-dispatch strategy of the wind-storage power plant,and realize the field-level online decision-making independent of prediction.Firstly,Lyapunov optimization was used to construct the basic strategy to obtain a good but not necessarily local optimal strategy.Then,samples were pre-generated by the basic strategy to initialize the experience base and improve the search efficiency.Further,DDPG algorithm with expert mechanism was applied to train the locally optimal self-scheduling strategy.Case study shows that compared with the basic dispatch strategy and the classical DDPG,the proposed method can effectively improve the average revenue of the wind-storage power plant.

关 键 词：风储联合电站 实时自调度 Lyapunov优化 深度确定性策略梯度(DDPG) 

分 类 号：TM614[电气工程—电力系统及自动化]