Abstract
User electricity data contains the characteristics of residential users' electricity consumption behavior. In order to help power companies formulate demand response plans and time of use electricity prices, and better extract electricity consumption behavior characteristics, this paper proposes an electricity consumption behavior classification model based on sparse denoising autoencoder (SDAE) feature dimensionality reduction and K-means clustering. Firstly, sparse denoising autoencoder is used to learn features, and K-means clustering is used for classification. Visualize the classification results using the t-distributed stochastic neighbor embedding (t-SNE) method, calculate typical user curves using Gaussian distance weighting, and analyze the characteristics of the electricity consumption curve. The effectiveness of the proposed method was verified by calculating and comparing the clustering indicators of other common dimensionality reduction methods.