In order to adapt to the market demand in the power big data environment, it is particularly critical to accurately analyze the power load characteristics and adjustable capabilities of users. In this piece of writing, a new comprehensive clustering method is proposed, which can classify the power consumption behavior according to the load attribute and its adjustability. Firstly, stacked denoising auto-encoder (SDAE) is used to extract features from users’ daily electricity consumption data. Secondly, in order to carry out the cluster analysis accurately, it is necessary to consider the various influencing factors of the user’s electricity consumption behavior, this paper combines the DTW (Dynamic Time Warping, DTW) algorithm with the K-shape algorithm to construct a user clustering model with similar characteristics of electricity consumption patterns, effectively identifying high-dimensional electricity consumption data and accurately extracting user electricity consumption information. Lastly, the selection of BDG2 load data and related variables influencing user consumption behavior validates the validity and applicability of the suggested clustering approach, and it is also proved that this method has generalization ability in multiple scenarios.