水力发电学报2026，Vol.45Issue(4)：59-72,14.DOI:10.11660/slfdxb.20260405
深度强化学习驱动的水光储互补系统优化调度
Deep reinforcement learning-driven optimal scheduling for hydro-photovoltaic-storage complementary systems
摘要
Abstract
The independent operation of hydropower,photovoltaic(PV)and energy storage stations is constrained by transmission channel capacity,leading to frequent curtailment of water and PV power,which limits the clean energy absorption capacity of power grids.To address such an issue,this paper describes an optimal scheduling method for hydro-PV-storage complementary systems based on the Asynchronous Advantage Actor-Critic(A3C)algorithm,which is applicable to large-scale hydro-PV-storage coordinated operation scenarios.First,an operational scenario of hydro-PV-storage stations is constructed,and an optimal scheduling model is built based on the short-term versus medium-and long-term complementary guidance mechanism.Then,for a hydro-PV-storage complementary system,we transform its optimal scheduling problem into a Markov decision process,and achieve efficient strategy exploration and learning via deep reinforcement learning algorithms.Finally,we validate the method through application to such a system in the Yarkant River basin,Xinjiang.Results show the A3C algorithm stably converges to high reward values and improves system-absorbed electricity significantly with its computational cost notably lower than other algorithms,demonstrating its promising practical application value.关键词
水光储系统/互补机制/深度强化学习/叶尔羌河流域/异步优势动作评价算法/马尔科夫决策过程Key words
hydro-photovoltaic-storage system/complementary mechanism/deep reinforcement learning/Yarkant River basin/asynchronous advantage actor-critic algorithm/Markov decision process分类
能源科技引用本文复制引用
向聪,黄显峰,李俊臣,周士浩,方国华,周论..深度强化学习驱动的水光储互补系统优化调度[J].水力发电学报,2026,45(4):59-72,14.基金项目
国家自然科学基金项目(52179012) （52179012）
中核集团科技项目(XJYH-2024-DL-QT011) （XJYH-2024-DL-QT011）
