Taking the rainstorm process in Sichuan Province from 20∶00 BST on 14 to 20∶00 BST on 15 July, 2021 as an example, this paper quantitatively analyzed the accuracy of five models (CMA-GFS, CMA-MESO, CMA-EPS, EC and EC-EPS) in forecasting the structure, intensity and location of rain bands by using the SAL test method, and further evaluates the spatial forecasting performance of each model in the main flood season (July) of Sichuan Province, and analyzed the accuracy of heavy rainfall process forecasting. Results show that: (1) the SAL method can intuitively provide forecasters with the prediction effect of numerical models on the structure, intensity and location of rainfall. (2) The EC model has the highest accuracy for rainstorm forecasting, and CMA-MESO has good reference value for the location forecasting of rain belt. (3) In the actual forecasting business,the heavy precipitation can adjust the heavy precipitation location refer to the CMA-MESO model, and the precipitation level can be corrected.