Intelligent distribution network situation awareness is an important foundation for the reliable, economic and safe operation of the distribution system. The scale and type of its data are growing rapidly, showing typical power big data characteristics. The method firstly uses moving and dynamic time windows to further mine the typical characteristics of various events, and then performs hierarchical labeling according to region selection and event types, reducing the computational dimension of machine learning algorithms and improving computational efficiency. Then this paper designs three types of classifiers and compares them with the other two types of classifiers. Through the test data of the example, it is concluded that the performance indicators of the proposed classifier are excellent.