To improve disaster prevention and the mitigation ability of power grid companies, taking into account the power grid, meteorology and geography, a power outage spatial prediction method in the case of typhoon disasters is proposed from the perspective of statistical learning. First, data are collected in a 1 km×1 km grid, and standardized, classified with variable one-hot encoding and screening, and construction features are processed as input data for the model. Secondly, algorithms such as random forest, gradient boosting decision tree, adaptive boosting, K-nearest neighbor, support vector machine, extreme tree, decision tree, and XGBoost are selected to construct an outage space prediction model using Stacking integration technology. Finally, a city in Guangdong Province is taken as the research object to verify the performance of the model: the prediction accuracy rate for typhoon "Mujigae" is 0.777 6, and the recall rate is 0.914 0. The results verify the feasibility and effectiveness of the model in predicting power outage space in the case of typhoon disasters. This work is supported by the University-Industry Collaborative Education Program of the Ministry of Education (No. 201902056044).