Load clustering can not only provide high-quality data for load forecasting, but also help carry out user behavior analysis. In order to meet the challenge of processing massive data, a dimension reduction and improved K-means clustering algorithm based on daily load indicators is proposed. Firstly, the original high-dimensional load data is converted into low-dimensional data by establishing a daily load indicator. Then the distributed K-means algorithm improved by the entropy weight method is used to cluster the low-dimensional data in order to discover hidden load types. Finally, according to the obtained typical load, the power consumption law is analyzed, and it is matched with the actual user type, and the four main power consumption laws are summarized.