Improved Load Curve Spectrum Clustering Method Considering Load Trend Self-correlation

Abstract

A large amount of electricity consumption data are accumulated via smart meter, and the load patterns of power users can be extracted via clustering algorithm. It is an important basis for power services such as load forecasting, time-of-use price, load control and electrical anomaly detection. For the problem of load curve clustering, the traditional algorithm has the disadvantages of low accuracy and poor stability, so it can not consider the characteristics of load trend. Firstly, the discrete wavelet transform is used to map the load curve to the frequency domain space, and the low-frequency characteristics of the curve are taken as the overall load trend through approximation method, and the shape features of the curve are extracted and the dimension data is reduced. Then, considering the morphological characteristics of the curve, the Euclidean distance and autocorrelation are used as the similarity measure, and then the spectral clustering is used to cluster. Finally, spectral clustering is used to cluster. The results show that the proposed method has greater advantages in clustering validity and stability compared with traditional methods in standard data set and measured smart meter data, and the clustering results can provide effective reference for business development such as time-of-use price.