Charging Load Classification Based on Improved CEEMDAN Noise Reduction and k-means++

The classification of charging load is conducive to mastering the charging behavior of users, optimizing the layout of charging infrastructure, and promoting the transformation of the consumer side structure of high energy consumption industrial parks. In this paper, entropy weight method is combined with traditional CEEMDAN to denoise and extract preliminary fault feature signals. Further, the signal is decomposed and filtered twice, and the Savitzky Golay smoothing filter is used to reconstruct the signal to reduce the noise of the charging load signal. By comprehensively calculating the sum of squares of errors, average profile coefficient and Davies Bouldin index, the process of selecting cluster number of K-means++is optimized, and the peak valley classification of charging load is realized. By analyzing the charging data of a charging station, the effectiveness of the improved CEEMDAN and K-means++methods proposed in this paper is verified.