Research And Application In Electrical Load Pattern Classification Oriented Distributed Clustering Algorithm

Electrical load clustering has become more and more important in modern society,and has many applications in load forecasting,abnormal electricity usage detection and so on.Nowadays the clustering of electrical customers’ load profile mainly faces two problems.One problem is that as the popularity of the load collecting tools,more and more load data about the electrical consumers is gathered.As known so far,the first-tier cities are facing a quantity of millions of load data to collect in its peak period of electrical usage.It is very hard for traditional one-machine clustering algorithms to deal with that large amount of data properly and timely,so distributed clustering algorithms that fit with the characteristic of electrical consumers are needed.Another problem stand in the way is that the most popular clustering algorithm used in electrical user load classification is the Fuzzy C-Means(FCM)clustering algorithm,one of the weaknesses of Fuzzy C Means clustering algorithm is that it tends to equal the sizes of all the clusters,so it is not suitable for dealing with circumstances where cluster sizes vary greatly.While for electrical consumers,it is quite normal to have clusters of different sizes ranges enormously.To solve the two problems mentioned above,in this thesis we studied and modified relational clustering algorithms to solve the weakness of FCM algorithm of not suitable for clusters of different sizes,and then we researched relevant distributed clustering algorithms so that the modified clustering algorithm can be executed in decentralized circumstances.We also researched some papers about how to cluster electrical consumers’ load curves and using the representative load curves in abnormal electricity usage detection.The main work of this thesis is as follows:1.This thesis modified the csi-FCM algorithm,during our study,we found that althrough csi-FCM algorithm can solve the size-sensitive problem in some degree,it also had some weaknesses.It is sensitive to cluster initilizations and relative distance between different clusters.We solved the problem by enhancing the importance of the pure data of each cluster so that the modified algorithm can be more robust.2.This thesis used the modified leaders algorithm to better represent the raw dataset to gain the aim of data reduction.The modified algorithm was decentralized by collecting all the representatives from all the local nodes to implement global clustering.3.We used the new distributed clustering algorithm to cluster electrical consumers’ load data and used the representative load curves of each cluster to detect abnormal electricity usage,and got higher accuracy with the new distributed algorithm than FCM.