Virtual power plant flexible resource aggregation model based on a two-stage classification method

FAN Dawei; HU Wei; TIAN Zhan; DU Puliang; DAI Jiazhi

Abstract：With the popularity of virtual power plants,the flexible resources of distributed power plants are aggregated into one through virtual power plants to participate in power grid scheduling.Analyzing the characteristics of various flexible resources and optimizing various flexible resource scheduling schemes are the key paths to realize the stable operation of power grids.A two-stage classification algorithm for virtual power plant flexible resources was proposed to solve the problems of one-sided classification results,repeated classification of classifier and unequal classification of load types in machine learning load classification.In the first stage,spectral clustering algorithm and long short-term memory(LSTM)neural algorithm are combined to classify the flexible resources of virtual power plants.To solve the problem of low learning efficiency of machine learning unit classifier,a unit classifier based on LSTM method is proposed.Then,in order to solve the classification unevenness problem of base classifier,an optimal selection method based on risk minimization loss is proposed.Finally,the problem of uneven load classification is solved by improved Gaussian oversampling method.In the second stage,according to the classification results of the flexible resource load curve of the virtual power plant,the flexible resource category of the virtual power plant is constructed by the flexible resource load curve.Finally,the effectiveness of the proposed method is verified by virtual power plant load data.