In view of the complex and changeable power consumption behavior of users under the background of power big data and the difficulty in analysis, a classification and analysis method of power consumption behavior of users based on undercomplete auto-encoder is proposed.Firstly, the data of intelligent electricity meters are encoded by an undercomplete auto-encoder to extract the features of the original data, and the back-propagation (BP)neural network is used to classify and analyze the user′s electricity consumption behavior.Then, the optimal coding ratio is selected, and the typical user electricity characteristics are taken as the input of the neural network to improve the classification accuracy.Finally, a simulation experiment is carried out on smart meters in Ireland data sets, compared with directly using the BP neural network analysis, the proposed method not only can improve the accuracy of detection, help electric power company to better grasp the power law of auxiliary demand response, but also can significantly reduce the running time of the algorithm.