Research On Classification Of Electricity Customers By Real Load Curves

Clustering load curves is an important task in obtaining the typical load profiles (TLPs) of electricity customers and grouping them into classes according to their load characteristics, it has theoretical and practical significance for load forecasting, load control, detection of electricity behavior irregularities and even improving tariff structures and rates and developing new marketing strategies .Current, there are many existing clustering methods on the load curve, which has its own advantages and disadvantages. In this paper, popular clustering algorithms are firstly studied and analyzed, including the k-means, the average and Ward hierarchical methods, fuzzy c-means (Fuzzy c-mean,FCM), the self-organizing map (Self-organizing Feature Maps, SOM), also algorithms such as the Gaussian mixture model clustering algorithm (Gaussian Mixture Model,GMM), extreme learning machine (Extreme Learning Machine,ELM), support vector machine (Support Vector Machine,SVM) are investagted. For the popular clustering algorithms, they are compared to obtain the optimum clustering algorithms by internationally recognized data set IRIS, common curve clustering index, the real load curve of electricity customers. Then these optimum clustering algorithms are used to obtain the typical load profiles (TLPs) and achieve the load pattern classification of electricity customers, and the compositions in terms of current tariffs and industrial classification of national economic activities in each group are presented, the difference between them is important for for developing power market in demand side. Finally, the selected optimum algorithms are used to identify behavior irregularities of electricity customers, which will assistant the inspectors to improve the hit rate. This study shows that Load pattern analysis by clustering load curves has good application prospect in the smart distribution system.