Abstract: In order to improve the distribution network equipment utilization, a method for optimizing the distribution network equipment utilization, taking into account the deep learning load classification and its demand response, is proposed in this paper. The LSTM self-encoder in deep learning is used to perform feature layer extraction and dimensionality reduction on the typical daily load curve set through training. The improved K-means load clustering algorithm is used to perform cluster analysis on feature sequences. Based on the results of load classification, an optimization model of distribution network equipment utilization that takes into account the UOT demand response of classified loads is established. The goal of the model is maximizing equipment utilization and user satisfaction. The analysis results of examples show that the proposed method can better stimulate the potential of various load demand responses,guide users to cut peaks and fill valleys through classified load demand responses; and effectively improve the distribution network equipment utilization, and the balance between the system equipment utilization and user satisfaction can be realized through rationally setting the equipment utilization and user satisfaction goal weights.