The Research Of Electricity Customer Load Pattern Based On Clustering Ensemble Approaches

With the rapid development of smart grid,many intelligent measurement devices are widely used in power network to monitor,control and predict the use of power.Grasp the load pattern effectively can help power companies understand the behavior of the users,subdivide the customer groups and improve the reliability of power system operation.In addition,it can help users to energy-saving reformation and improve the economic benefit and so on.Therefore,the research of load pattern of power users is of great significance.In this paper,a new method of pattern classification based on the technology of factor analysis and clustering ensemble approaches is presented.The clustering results achieved by electricity consumption data show that it can effectively improve the accuracy of clustering and reduce computation complexity.In this paper,the technologies and methods of extracting power user load pattern are introduced in detail,which provides a foundation for the proposed clustering ensemble approaches.The most commonly used load pattern extraction method is the clustering method.Most of the power researchers focus on the selection and optimization of clustering methods,but the influence of the key task(data preprocessing)of the early extraction of load ptttern on the clustering effect is seldom mentioned.In this paper,the influence of the data pretreatment methods on the clustering results is studied,and the adaptive relationship of clustering method and data standardization method is obtained.Secondly,in view of existing situation that large data sets which the load dimension is high lead to clustering efficiency is low,several kinds of commonly used data set dimension reduction methods are analyzed,such as self-organization mapping,sammon mapping,factor analysis.Considering the efficiency of clustering and the accuracy of clustering,factor analysis method is used for dimension data set in this paper.Factor analysis is a kind of feature extraction and data compression methods through the linear combination of the original variables.Then,a few unrelated variables are generated and these variables in the reaction of original variable information as much as possible.In this paper,experiment shows that the load dimension,the data storage space and the computation time of the algorithm is reduced on the premise of the original information as much as possible.Finally,to solve the problem about the clustering results of traditional single clustering algorithm is unsteady and the reliability of load pattern extraction is low,clustering ensemble technology is applied in this paper to extract electricity user load pattern.Then,four clustering algorithms:K-means,fuzzy C mean,cohesive hierarchical clustering Ward method,and self-organizing mapping neural network are used to get different cluster members.The clustering members are combined with the Co-association Matrix and the clustering result is better than single clustering algorithm.