In the complex terrain of Sichuan region, although the frequency of thunderstorm gale is relatively low, the impact is significant, and there are few existing objective forecast products with low time resolution. In order to further improve the accuracy of thunderstorm gale forecasting under complex terrain in Sichuan, this article comprehensively considers terrain factors, model physical quantity factors, and time factors. According to the altitude, Sichuan is divided into high-altitude and low altitude areas. Based on three machine learning methods: random forest, adaptive boosting, and extreme random tree, a thunderstorm gale prediction model was constructed by using data from 2018 to 2021, makes a forecast for 2022 and obtains a 3-hour thunderstorm gale potential forecast, and then, using the climate background, the 3-hour forecast time is scaled down to 1-hour to form a 0-12 hour hourly thunderstorm gale forecast, and the forecasting effect is tested. The results showed that, the adaptive boosting method 3-hour thunderstorm gale forecast has the best effect, long term and individual case tests showed that the 0-12 hour thunderstorm gale forecast product obtained from adaptive boosting method is superior to National Meteorological Centre (NMC) forecast product, the TS score increased from 1.04 % to 5.95 %, and the false alarm rate decreased from 98.8 % to 80.8 %, indicating high business application value.