Power network reliability early warning method based on node importance

The invention relates to a power network reliability early warning method based on node importance. Comprising the following steps: carrying out graph attention network topology importance evaluation fusing node features, carrying out feature aggregation on low-dimensional features of nodes through a graph attention network model, and outputting the topology physical importance degree of each node after learning and marking a training set; calculating the business importance degree of the node by counting the type and the number of businesses borne by each node; a fuzzy analytic hierarchy process is used to determine weight coefficients of the two, and comprehensive importance of the nodes is considered; and finally, according to the evaluated node importance degree, collecting invalid node information in the network, calculating a network reliability early warning value and judging whether to give an alarm or not. According to the method, the topology importance and the service importance of the nodes in the