Abstract:  When there are many kinds of load curves, the efficiency of the traditional clustering method is not high in user load classification, which hinders the application of clustering method in the big data analysis of power load. This paper proposes a ‘Evolution’ principal component analysis (PCA) method. Firstly, we adopt PCA to reduce the load matrix dimensionality of users； then, proposes the classification rules based on Euclidean distance, on the basis of PCA. Taking the actual load of users in a certain area as an example, all kinds of user curve shapes are fitted by cosine similarity theorem, which verifies the effectiveness of the proposed algorithm. Compared with traditional load curve classification method, it is showed that the ‘Evolution’-based PCA can improve the classification efficiency of load curve. On the basis of load curve classification, compared with the local overall load curve, the user is divided into 4 categories: peak electricity users, part meeting peak electricity users, a few meeting peak electricity and abnormal electric type. The analysis results show the effectiveness and practicability of the load classification based on ‘Evolution’ PCA. The proposed load classification method can be more effective in the classification of user behaviour, so as to establish the dynamic electricity price for all kinds of users, which can be the basis for the development of smart grid related value-added services.

Key words:   smart grid, principal component analysis(PCA), user classification, behavior analysis