Power User Load Pattern Recognition Based On Improved Bisecting K-means And Ensemble Learning

With the deepening of the reform of the power market and the construction of smart grid,the businesses carried by the power market are more and more complex and diverse.Meanwhile,a large number of intelligent measuring devices are put into use,resulting in the continuous expansion of the scale of the power load data network.Using effective data mining technology to analyze and utilize these data is of great significance to the stable operation of power grid and the healthy development of power market.Load pattern recognition by studying the characteristics of different load data,different types of load patterns are divided,and corresponding strategies are formulated for different load patterns,which has important reference value and significance for the load analysis and planning,demand control and response,and load prediction of power system.The research of load pattern recognition mainly focuses on cluster analysis and classification of load curve.Firstly,cluster analysis is carried out on the data sets in the load database to obtain the data characteristics of various typical loads,and different processing methods are formulated for different types of loads.Then,choose different kinds of load data as training samples,training the ability to identify the load type of classifier,through the trained classifier measurement equipment in the power network can be collected by the new type of unknown load data classification recognition,and divided them into one type of load and thus can be used for the load approach to deal with the new collected unknown type load.In this paper,the existing problems in traditional load clustering and classification methods are studied and improved.The main contents are as follows:(1)In order to solve the difficulty of calculation and storage in the processing of high-dimensional load data by traditional clustering algorithm,t-SNE dimensionality reduction technology is applied to the processing of load data,which has a better effect than the commonly used PCA dimensionality reduction technology.(2)The K-means algorithm commonly used in load clustering algorithm has two main problems:one is that it depends on the choice of the initial center of mass;the other is that the number of clusters should be determined in advance.This paper used bisecting K-means algorithm based on t-SNE and GSA elbow criterion,which solves two major problems of traditional k-means algorithm and significantly improves the clustering quality.(3)Generally,the methods of load classification and recognition are used to train a single classifier through a large number of samples to improve the accuracy,but the learning ability of a single classifier is limited after all.In this paper,Adaboost integrated learning algorithm is combined with BP neural network,and multiple classifier integrated learning is combined into a high-precision classifier to improve the classification accuracy.(4)Through the method of load pattern recognition,the characteristics of typical load patterns are studied,and the load characteristics of typical industries in different load patterns are analyzed.