Abstract: Based on hourly precipitation observation data at 10,189 meteorological stations from April to September 2018 in Northwest China and ECMWF prediction data with 0.25°×0.25° resolution,a precipitation prediction model for Hexi Corridor was established by using V-3θ diagram and Keras method. The prediction skill of the precipitation prediction model was examined by three methods: "loss function of classification test","TS test" and "root mean square error of fitting (RMSE) test". The results show that the two neural network frameworks established by Keras method,using "K-fold crossover" and logistic regression method,can improve the rationality of feature relation,reduce the error of feature quantity,and make the model prediction results more reliable. The V-3θ diagram was derived,and the types of characteristic values were increased,which reduced the subjective identification deviation of precipitation prediction model in Hexi Corridor,and realized the objective quantification of precipitation vertical structure prediction. After being tested by three methods,it is found that the prediction result of the model in daytime is better than that at night,and the prediction result in the range of 12 hours and 18 hours is the most consistent with the actual value. Through the comparison of individual cases,it is found that the model can accurately predict the occurrence time of precipitation,the main precipitation period,the regional range of precipitation and the intensity of precipitation center,and has a strong ability to predict heavy precipitation weather.