Hybrid Features based K-means Clustering Algorithm for use in Electricity Customer Load Pattern Analysis

摘要

In the power system, the load pattern of users is important for electricity price formulation, abnormal electricity consumption detection and load forecasting. Accurately identifying the load patterns effectively improve the overall load forecasting performance of the power grid. However, due to many factors affecting the load, identifying load patterns is not a simple job. In this paper, the features of load data are extracted and different load patterns of users are identified. In the pattern classification, K-means clustering is used. In feature extraction, PCA and prior knowledge were respectively used for dimensionality reduction to extract new features in the past. Combining the advantages of PCA and prior knowledge, a new method, namely, weighted combination of the obtained features from two former methods, is adopted to extract new features and identify the user's load pattern. In the similarity measure, the variance of each feature is taken as its weight, and the clustering result is superior to that of PCA and prior knowledge. In the numerical simulation, it is found that the clustering results using the new method in this paper are better than those using only PCA to reduce dimensionality or using prior knowledge to reduce dimensionality. In the experiment, the users in a certain area were divided into four categories according to the clustering result of load data: peak electricity type, partial peak electricity type I, partial peak electricity type II and abnormal electricity type.