Abstract: Using radar reflectivity data from Sichuan Provincial Meteorological Service (2016—2020, May—September) and Tianfu International Airport radar reflectivity data (2021—2024, May—September), combined with concurrent ERA5 reanalysis data. The LASSO algorithm was employed for physical parameter screening. By evaluating the modeling results of various ensemble algorithms, a severe convection potential forecast model for Tianfu Airport was constructed. The results indicate that the model established using the LightGBM algorithm achieved the highest F2 score of 0.417. Key physical parameters influencing model performance include sea level pressure, 925 hPa dewpoint divergence, 700 hPa meridional wind, 500 hPa vorticity, and CAPE value. The average values of these parameters during severe convection events show significant differences compared to their averages over the summer months (May—September), reflecting the characteristics of severe convection weather at Tianfu International Airport and holding statistical significance. The construction of this model provides a methodological approach for physical parameter screening and demonstrates potential application prospects in meteorological forecasting.