Abstract: The prediction of the damaged number of towers in typhoon-affected areas is of great significance to improving the disaster prevention and mitigation capacity of power grid. From a perspective of brand-new statistical learning, based on the original samples of multi-source heterogeneous data such as meteorological data, power grid data, and geographical data, we pre-process the data such as classification variable processing, feature screening, standardization, etc. Taking the damaged number of towers in the grid as the response variable, we construct the prediction model of damaged number of towers under typhoon disasters based on the gradient lifting decision tree algorithm, and predict the damage quantity of 10 kV tower in distribution network under typhoon disasters. The model is trained, tested and verified by using the data samples of Xuwen County, Guangdong Province, China. The simulation results show that the proposed method can be adopted to accurately predict the number of damaged distribution towers under typhoon disasters, and provide auxiliary decision-making reference for pre-disaster risk quantitative assessment and emergency plan-making.