摘要
针对随机动态规划在求解风电—抽蓄联合系统日随机优化时出现的维数灾问题,提出采用强化学习的SARSA算法来解决。首先分析了风电出力随机性并采用Beta分布来表示风电出力的概率分布;然后建立了风蓄联合系统实际出力与计划出力偏差平方最小为目标函数的日随机优化模型;最后说明利用SARSA算法求解该问题的步骤。算例应用结果表明,利用SARSA算法求解该问题需迭代一定次数才收敛,且算法的学习率随迭代次数增加而减小时可加快算法收敛速度;将SARSA算法与随机动态规划算法相比,在优化结果接近的情况下,SARSA算法计算时间减少约35%,该算法为解决随机多能互补问题提供了新思路。
In order to solve the dimension disaster problem of stochastic dynamic programming in the daily stochastic optimization of wind-power pumped storage joint system,the SARSA algorithm is proposed based on reinforcement learning.Firstly,the randomness of wind power output was analyzed and Beta distribution was adopted to represent the probability distribution of wind power output.Secondly,a daily stochastic optimization model with the minimum square deviation of the actual output and planned output of the combined wind storage system as the objective function was established.Finally,the steps of using SARSA algorithm to solve the problem were explained.The example shows that the SARSA algorithm needs a certain number of iterations to solve the problem before convergence,and the learning rate of the algorithm decreases as the number of iterations increases,which can accelerate the convergence speed of the algorithm.Compared the SARSA algorithm with the stochastic dynamic programming algorithm,when the optimization results were close to each other,the calculation time of the SARSA algorithm was reduced by about 35%.This algorithm provides a new way to solve the stochastic multi-energy complementary problem.
作者
李文武
郑凯新
刘江鹏
贺中豪
LI Wen-wu;ZHENG Kai-xin;LIU Jiang-peng;HE Zhong-hao(College of Electrical Engineering&New Energy,China Three Gorges University,Yichang 443002,China;Hubei Key Laboratory of Cascaded Hydropower Stations Operation&Control,China Three Gorges University,Yichang 443002,China;State Grid Hubei Electric Power Co.LTD.,Yichang Power Supply Company,Yichang 443002,China)
出处
《水电能源科学》
北大核心
2020年第11期72-76,共5页
Water Resources and Power
基金
湖北省技术创新项目(重点项目)(2017AAA132)
梯级水电站运行与控制湖北省重点实验室(三峡大学)开放基金(2019KJX08)。
关键词
风蓄随机优化调度
强化学习
SARSA算法
学习率
wind-storage stochastic optimal scheduling
reinforcement learning
SARSA algorithm
learning rate
