Feature extraction and evaluation of electricity load data with high precision

This paper summarizes the characteristics of the electricity load data collected every 15 minutes of power users. The single-day load data of single power user is taken as a 96-length vector, and the single-day data of n users is taken as a 96-column set. Single-user monthly data and annual data are described as a 30 (31) × 96 matrix and a 365 (366) × 96 matrix respectively. By comparing the wavelet transforms of multiple wavelet basis functions, the Daubechies 4 wavelet basis function is chosen as the wavelet transform basis function of the electricity load data according to the Shannon entropy of the obtained wavelet coefficients. The difference between the inverse wavelet transformed data and the original data is compared under various characteristic wavelet coefficients. Besides, the number of the characteristic wavelet coefficients and the error of the wavelet transform are determined. The results show that there is a high redundancy in the original power load data, and the storage space can be greatly compressed by wavelet transform to achieve feature extraction and data desensitization. Therefore, the feature extraction technology of power load data based on wavelet transform in this paper has potential application value and popularization value, and can produce high economic efficiency.