Base station KPI prediction based on multi-graph convolutional network

The invention relates to base station KPI (Key Performance Indicator) prediction based on a multi-graph convolutional network. The method for predicting the KPI of the base station comprises the following steps: generating initial KPI data of the base station based on historical KPI data of the base station; inputting the initial KPI data into a graph convolutional network based on a neighbor graph, a graph convolutional network based on a similar graph and a graph convolutional network based on an interworking graph so as to respectively extract neighborhood features, scene features and mobility features; wherein the proximity graph, the similar graph and the intercommunication graph respectively indicate a proximity relation, scene similarity and intercommunication among a group of base stations including the base station; fusing the extracted neighborhood features, scene features and mobility features to obtain fused features; and calculating a prediction value of the KPI index based on the fusion feature.