电力建设2025，Vol.46Issue(9)：84-97,14.DOI:10.12204/j.issn.1000-7229.2025.09.007
基于条件风险强化学习的梯级水光蓄联合优化调度
Condition-Based Risk Reinforcement Learning for Joint Optimal Scheduling of Cascade Hydropower and Solar-Powered Reservoirs
摘要
Abstract
[Objective]Owing to the advancement of the dual-carbon strategy,hydropower developers are actively promoting the pumped-storage retrofitting of hydroelectric facilities while accelerating the regional construction of photovoltaic(PV)and other renewable-energy sources,thus gradually forming a scenario with high-penetration renewables and large-scale pumped storage.However,under the dual uncertainties of renewable-energy output and interval water inflow,the system operational coordination becomes more complex,particularly because scheduling methods for cascade hydropower systems that consider interval inflow uncertainty are rarely investigated.[Methods]This study proposes a conditional risk-aware reinforcement-learning-based optimal scheduling method for cascade hydro-PV-pumped storage systems.First,the"Informer"deep neural network is employed to predict basin-interval water inflow,where inflow uncertainty is transformed into flexibility supply indicators.Subsequently,risk theory is integrated to quantify flexibility deficits using the conditional-value-at-risk measure.Finally,an improved risk-managing proximal policy optimization(RM-PPO)reinforcement-learning algorithm is developed to derive optimized scheduling strategies.[Results]Validation using an actual cascade hydro-PV-pumped storage base in China shows that the proposed forecasting method reduces the MSE and MAE by 18.9%and 58.8%,respectively,compared with conventional time-series approaches,with a 78.5%capture rate for peak events.The RM-PPO scheduling algorithm achieves flexible system regulation through pumped-storage plants coordinated with PVs for surplus-energy absorption,whereas conventional hydropower stations synergize with PVs for source-load dynamic matching.[Conclusions]The RM-PPO-based scheduling strategy effectively reduces costs and controls operational risks while maintaining system flexibility,thus promoting renewable-energy accommodation and enhancing hydropower-utilization efficiency.关键词
梯级水光蓄/条件风险/强化学习/调度优化Key words
cascade hydropower-photovoltaic storage/conditional risk/reinforcement learning/scheduling optimization分类
信息技术与安全科学引用本文复制引用
陈实,唐国登,刘艺洪,许刘超,朱钰杰,周毅,李华强,臧天磊..基于条件风险强化学习的梯级水光蓄联合优化调度[J].电力建设,2025,46(9):84-97,14.基金项目
国家自然科学基金资助项目(52377115)This work is supported by the National Natural Science Foundation of China(NSFC)(No.52377115). （52377115）
