Abstract With the integration of renewable energy and microclimate-sensitive loads, secure and economic power system operation is becoming an increasingly important and complex problem. Therefore, based on big data from power systems and meteorological systems, a deep spatial-temporal data-driven model is proposed to predict and detect power system security weak spots during a future period. First, microclimates are considered in the proposed model. Then, a deep neural network structure is designed to extract deep features layer by layer for security weak spot detection. Furthermore, model simplification and parallelism as well as data parallelism are applied. Finally, the proposed model is evaluated based on the Guangdong Power Grid in China. The simulation results demonstrate that (1) power system security weak spots have spatial-temporal and microclimate-sensitive characteristics; (2) the deep model considering microclimates can greatly improve the task accuracy of online applications; and (3) simplification and parallelism can significantly enhance the training efficiency.