The invention relates to an electric power system safety early warning method based on electric power-microclimate spatio-temporal data, and belongs to the technical field of electric power system safety analysis. A deep neural network model is adopted, and a new technical scheme is provided for safety early warning of the power system under the condition of considering micrometeorological factors. According to the method, on one hand, a data driving model is used, so that the problem of low calculation efficiency of a model driving method in the problems of high-dimensional uncertainty and multi-system combinatorial analysis is solved. On the other hand, the microclimate characteristic data, the power system characteristic data and the space-time characteristics thereof are comprehensively considered, and the relation between the power system and the microclimate factors is realized. The method can be applied to the field of safety early warning of the power system, improves the accuracy of safety early warning, provides a scheduling basis for operation scheduling personnel of the power system, and facilitates the improvement of the scheduling safety of the power system.