权衡生态和发电目标的梯级水库强化学习模型及其应用  

Reinforcement Learning Model and Its Application for Balancing Ecological and Power Generation Objectives of Cascaded Reservoirs

作　　者：王一聪 陈明洪[1] WANG Yi-cong;CHEN Ming-hong(College of Water Resources and Civil Engineering,China Agricultural University,Beijing 100083,China)

机构地区：[1]中国农业大学水利与土木工程学院,北京100083

出　　处：《水电能源科学》2024年第2期188-191,206,共5页Water Resources and Power

基　　金：国家重点研发计划(2022YFC3201804,2016YFC0402506)。

摘　　要：为更好地发挥经济及生态效益,建立了基于强化学习算法的梯级水库优化调度模型,以周尺度小浪底天然入库流量过程为基础,探讨梯级水库发电与生态目标之间的权衡,并将该模型应用于小浪底-西霞院梯级水库中,分别探讨了不同调度方案下发电最优、生态最优和发电-生态权衡的优化调度策略。结果表明,以发电最优为目标时,梯级水库多年平均发电量比常规调度增加了3.62%~7.92%;以生态最优为目标时,平均生态保证率比常规调度增加了31.68%~33.66%。结果为梯级水库多目标优化调度提供了一种可行方法。To better leverage economic and ecological benefits,a reinforced learning algorithm-based model for optimal scheduling of cascaded reservoirs was developed to explore the trade-off between power generation and ecological objectives based on the natural inlet flow process of Xiaolangdi at the weekly scale.The model was applied to the Xiaolang-di-Xiaoxiyuan cascaded reservoirs,and the optimal scheduling strategies of power generation optimization,ecological optimization and power generation-ecological trade-off under different scheduling schemes were explored respectively.The results show that with the goal of optimal power generation,the average annual power generation of cascade reservoirs has increased by 3.62%-7.92%compared to conventional scheduling;When targeting ecological optimization,the average ecological guarantee rate increases by 31.68%-33.66% compared to conventional scheduling.Thus,it provides a feasible method for multi-objective optimal scheduling of cascaded reservoirs.

关 键 词：多目标调度 生态保证率 发电量 强化学习 小浪底水库 西霞院水库 

分 类 号：TV697.12[水利工程—水利水电工程]