Objective Meteorological disaster is one of the most important adverse factors that threaten the safe and stable operation of power grid. To enhance the meteorological disaster prevention capability of the power grid in complex terrain of Chongqing,

Method In this paper, the temporal and spatial distribution of lighting, wind and icing disasters were analyzed based on the weather-related power grid accidents and the ERA5 meteorological reanalysis data from 2014 to 2024 in Chongqing. Then the Support Vector Machine (SVM) and Random Forests (RF) algorithms were used to build a recognition model, and the accuracies were calculated.

Result The results show that the occurrence frequency of lightning disasters is characterized by double peaks in spring and summer, while wind and icing disasters are concentrated in spring and winter, respectively. Lightning and wind disasters are concentrated in Wuling and Daba Mountains, while icing mainly occurs in the southeast area with altitude over 800 m, which is also a multi-hazard superimposed high-incidence area.

Conclusion The identification accuracies of two machine learning algorithms for power grid disasters are compared in this paper, shows that the recognition abilities of RF algorithm for power grid disasters are better than that of SVM, for the accuracies of wind and icing disasters are more than 99%. Among the input characteristics of the model, the effect of wind, dynamic and temperature-humidity parameters for the recognition accuracies of lightning and icing disasters are equivalent, and which of the temperature-humidity parameters is slightly higher than that of the other two types of parameters. Comparing the recognition accuracy in different regions and seasons, it is found that the identification accuracies of the southeast and northeast is better than that of other regions, and the best seasons for the identification of lightning, wind and icing disasters are summer, spring and winter, respectively, which are all more than 95%.