Research On The Method Of Recognizing Specific User Load Patterns Based On K-means Clustering Algorithm

With the improvement of the intelligence level of the power grid,more and more intelligent measurement equipment has been invested in the power grid,and a large amount of complex load data has been accumulated.The State Grid corporation needs to process them,and tap the effective information to provide support for the normal operation of the power grid.The power grid system planning,load forecasting,demand side management,time-of-use electricity price,load modeling and other issues need to be pre-processed based on load clustering.The traditional method of classifying load data according to user types has been unable to meet the current needs of diversified power grid services.Therefore,a more scientific load classification method is urgently needed.According to the load data collected by smart meters throughout the day,daily load curve diagrams of power consumers can be drawn.Cluster analysis of them can obtain the power consumption characteristics and typical user daily load curves of the region,providing a scientific basis for reliable operation of the power system.It is of great significance to improve the utilization efficiency of power grid assets and save energy.Therefore,how to effectively and quickly cluster large-scale load data has become a hot topic at home and abroad.This paper mainly improves clustering and pattern recognition methods to study the classification of daily user load.Mainly carried out the following four aspects of work:1.Review the methods and research status of sorting data classification at home and abroad,research the characteristics of load data,and introduce the preprocessing of load data in detail,such as: detection and repair of outliers of load data,and normalization of load data.2.In order to reduce the dimension of load data,this paper introduces two methods:dimension reduction method based on load curve characteristic index and principal component analysis(PCA).The former extracts 10 features from each load curve for data dimensionality reduction.The latter selects the first 8-dimensional data with a cumulative contribution rate of 95%.3.An in-depth analysis of the K-means clustering algorithm is made,and two shortcomings of the clustering algorithm are pointed out,and two methods are proposed to optimize the K-means clustering algorithm.Based on the GSA elbow criterion,the K number of clusters is determined.For the selection of the initial clustering center,a density clustering algorithm is proposed.It also introduces the maximum and minimum distance weighted density method,which not only makes up for the shortcomings of the K-means clustering algorithm,but also solves the problem that the first optimization method is too dependent on the accuracy of the clustering.4.The K-means clustering algorithm optimized by two improved methods is used for complex load data clustering.The comparison is made from two aspects of time efficiency and clustering results.A more scientific optimization method is selected to optimize the K-means clustering algorithm,and the load data clustering results are analyzed.The innovation of this paper is to propose a weighted density method based on the maximum and minimum distance.The optimization algorithm has high calculation efficiency and stable clustering results.It also makes up for the problems of uncontrollable factors and unscientific processing flow in determining K value based on GSA elbow criterion.