Abstract: In recent years, the frequency of devastating typhoons has been on the rise. Typhoon “Talim” affected the power grid in Guangdong and impacted approximately a million users in 2023. We analyze the damage of Guangdong power grid and establish a prediction model for the destruction of transmission and distribution towers. Key features are identified to support disaster prevention. Firstly, we analyze the typhoon’s characteristics. It has “active convection in front of the typhoon, strong wind strength and wide precipitation range”, causing varying degrees of damage to transmission and distribution equipment. Then, using typhoon “Talim”, prediction models for predicting pole damage in the transmission and distribution network are established by Random Forest, Support Vector Machine, Gradient Boosting Decision Tree, and Neural Networks. The performance comparison is conducted before and after optimizing for imbalanced samples. Case studies indicate that Random Forest exhibits the greatest improvement after optimization, and considering both time and prediction quality indicators, the Gradient Decision Tree is the optimal algorithm.Finally, feature analysis and Shapley Additive Explanations are performed. The findings show that the maximum wind speed,temperature and the number of towers have significant influence on the prediction results. It is helpful to further understand the influence mechanism of typhoon on Guangdong power grid and provide reference for improving the ability of regional power grid to withstand complex natural disasters.