A Research And Implementation Of Multi-dimensional Long-sequence Forecasting Method For Large-scale Precipitation Forecasting

Large-scale precipitation forecast for nowcasting convective weather play an impor-tant role in the prevention of national meteorological disaster,especially,there is an urgent need for national agriculture,major social activities and disaster prevention.In recent years,Although the country has made great progress in strong convective weather nowcasting,there are many challenges in long range precipitation forecast.Deep learn-ing method is trying to solve the problem of Large-scale multidimensional long-sequence precipitation forecast.Strong convective cloud identification and movement prediction are the keys in Large-scale severe convective weather nowcasting.This paper relies on cetc14 research project.The following work is carried out to study and realize the methods of identifying and predicting the motion of strong convective clouds in the large-scale precipitation forecast:1)This paper analyzes the characteristics of large-scale precipitation prediction,stud-ies the principle of nowcasting in severe convective weather,and presents a method of large-scale precipitation prediction based on the recognition of multi-dimensional long-sequence of severe convective clouds and the prediction of movement of severe convective clouds.2)Analyzing model requirements data,and the data sets were processed by analyzing the national multi-site radar data,satellite data,and precipitable data.3)In view of the low accuracy and poor effect of the recognition of strong convective clouds in the large-scale satellite cloud image,a strong convective cloud recognition model based on image semantic segmentation is proposed.The deep learning neural network is more accurate than the traditional image recognition method and more suitable for the recognition model of strong convective clouds.4)For the prediction model of the movement of strong convective clouds with multi-dimensional long sequences,a prediction model of the movement of strong convective clouds based on stochastic adversarial video prediction model is proposed.The model can effectively improve the accuracy of large-scale precipitation forecast.At last,test accuracy of strong convective cloud identification model and prediction model of convective cloud motion.The test results of POD,FAR,CSI show that the model presented in this paper has a good effect on multi-dimensional long sequence prediction for large-scale precipitation forecast.