摘要
模型驱动的优化方法已无法解决风、光的间歇性和负荷的波动性给多能微电网系统的调度带来的问题,在此背景下提出了一种基于三维K均值聚类算法(K-means Clustering Algorithm,K-means)及深度确定性策略梯度(Improved Deep Deterministic Policy Gradient,IDDPG)算法的调度方法,以解决复杂环境下的随机性问题。首先按照风、光出力和负荷需求将微电网环境数据划分为三个维度并采用K-means对三维空间下的数据进行分类,然后将分类后的数据分别交互至DDPG中,同时构建综合能源微电网系统的数学模型,选取状态空间、调度策略和奖励函数,随后分别训练各类别的数据得出不同类别微电网数据下DDPG的参数,同时自适应改变动作探索范围,并根据微电网模型选取最优动作策略,最后将所提算法应用在某高校实际微电网算例中,并证明此算法无论在收敛性和给微电网带来的经济效益方面都优于DDPG和传统的调度方法。
Model-driven optimization methods can no longer solve the problems caused by the intermittency of wind and light and the fluctuation of load to the dispatch of multi-energy microgrid systems.In this context,a scheduling method based on three-dimensional K-means clustering algorithm(K-means)and improved deep deterministic policy gradient(IDDPG)algorithm were proposed to solve the problem of randomness in complex environments.Firstly,according to wind,solar output and load demand,the microgrid environmental data was divided into three dimensions,and K-means was used to classify the data in three-dimensional space.Then,the classified data was exchanged into DDPG,and a mathematical model of the integrated energy microgrid system was constructed.The state space,scheduling strategy and reward function were selected,and the parameters of DDPG under different types of microgrid data were obtained by training each category of data respectively.At the same time,the action exploration range was adaptively changed,and the optimal action strategy was selected according to the microgrid model.Finally,the proposed algorithm was applied to the actual microgrid study of a university,and it was proved that the algorithm was superior to DDPG and traditional dispatching methods in terms of convergence and economic benefits to microgrid.
作者
王珣
王钟
沈海华
肖勇
成贵学
蒋明喆
WANG Xun;WANG Zhong;SHEN Haihua;XIAO Yong;CHENG Guixue;JIANG Mingzhe(State Grid Huzhou Electric Power Company,Huzhou 313000,China;Shanghai Electric Power University,Shanghai 201306,China;State Grid Bengbu Electric Power Company,Bengbu 233090,China)
出处
《电工技术》
2023年第6期111-117,120,共8页
Electric Engineering
关键词
综合能源系统调度
深度确定性策略梯度
K均值聚类算法
随机性处理
integrated energy system dispatching
improved deep deterministic policy gradient
K-means clustering algorithm
randomness processing
