Abstract: Traditional forecasting methods are difficult to effectively capture severe convective weather rapidly evolving characteristics, and there is still a problem of insufficient forecasting accuracy in meteorological operations. Fully convolutional neural networks, due to their unique local perception characteristics, can effectively extract the spatiotemporal evolution features of small and medium-sized weather systems. Therefore, it is urgent to construct a prediction model based on fully convolutional neural networks to overcome the limitations of traditional methods. This article uses ground observations of severe convective weather from 2017 to 2021 and ECMWF (the European Center of Medium range weather forecasts) numerical model data as the training set, and 2022 as the test set. The FCN (Fully Convolutional Network) algorithm was adopted to construct three types of severe convective weather forecast models for hail, convective gust, and short-term heavy rainfall within 0-12 hours in Gansu Province. The model was applied in forecast business operations in 2023. The results show that, (1) In the training set of 2022, the model performed well, with an overall false positive rate(FNR) of only 16.6% for severe convective weather and non-severe convective weather. The average critical success index (CSI) for three types of severe convective weather is 25.8%, and the average hit rate (POD) remains above 65.2%, with short and strong weather forecasts showing the best results. (2) In the validation set of forecast business operations in 2023, the average CSI of the three types of severe convective weather is 24.3%, the average POD is 62.6%, and the average false alarm ratio (FAR) is 71.2%. On all datasets from 2017 to 2023, the CSI of short-term heavy rainfall was the highest, reaching 45.5%, while the POD for thunderstorm and strong winds exceeded 70%. The FAR results for hail and thunderstorm winds were similar, both exceeding 77%, while the average FAR for short and strong winds was the lowest. (3) From the hourly forecast of the model, hail performs best in the 4th, 8th, and 10th hours, thunderstorm winds perform best in the 6th hours, and short-term intensity forecasts perform best in the 2nd and 4th hours. Therefore, the FCN model constructed in this study performs well in strong convective weather forecasting, providing application prospects for future meteorological business automation.