面向保供需求的深度强化学习水-风-光短期优化调度研究OA

To balance the regulation capacity of hydropower stations and the supply reliability of hydro-wind-solar system,we proposed a short-term multi-objective hydro-wind-solar optimization dispatch model targeting supply assurance demand.Aiming at maximizing daily power generation and optimizing load profile matching,we employed SAC(soft actor-critic),a deep reinforcement learning algorithm,to solve the optimal scheduling model.The target curve feature decomposition method was introduced to de-couple the multiple objectives,and the hierarchical reward mechanism was designed to improve the strat-egy convergence performance.Taking the Pubugou,Shenxigou,Zhentouba(Pu-Shen-Zhen)cascade hydro-wind-solar system of the Dadu River Basin as a case study,we simulated the system operational states of typical days in flood season,normal season and dry season,under the scenarios of high and low output of wind and solar power.The simulation results were then compared with those of the conventional optimization algorithms,namely,progressive optimization algorithm(POA)and genetic algorithm(GA).The comparison shows that SAC exhibits excellent load tracking performance across all test scenari-os(with source-load correlation coefficient>99.9%),and achieves solutions with higher system power generation.It can achieve an average increase of 0.54%and 0.27%in cascade hydropower generation,compared with POA and GA.This result demonstrates that the maximum entropy exploration mechanism of SAC can effectively tap potential high-quality solutions,and balance the reliability of power supply as-surance and power generation,when the operational constraints are met.The proposed method can pro-vide a theoretical support and technical path for the intelligent optimal operation of hydro-wind-solar integrated systems.