Research On Meteorological Forecast Lattice Correction Algorithm Based On Spatio-temporal Sequence Network

Meteorology is closely related to human activities.The accuracy of weather forecasting greatly affects military,people’s livelihood,and economic fields.Extreme weather changes will destroy human living environment.The grid point meteorological element forecasting is a kind of classification of the area into grid points according to the established range,and the meteorological element forecast is carried out by grid points.At present,there is a forecasting service for meteorological elements of 5 km grid points.Fine-grained meteorological element grid point forecasting is a high-resolution forecast in grid point forecasting.The forecasting method usually refers to the current numerical forecasting,combined with the observation data,and the grid point data is corrected according to the experience of the practitioners.Make fine processing and correct errors.However,due to the lack of accuracy of traditional numerical forecasting in high-resolution short-term forecasting,the method of strengthening forecast accuracy is still based on manual correction,and how to improve the accuracy of short-term forecasting with the deployment of refined meteorological element grid forecasting services And the degree of refinement of refined forecasting is a significant research direction.This paper first introduces and analyzes the working mode and grid data of the refined grid point meteorological element forecasting correction.According to the spatial and time series characteristics of meteorological element changes,this paper introduces the ConvGRU feature extraction structure of the depth learning field to solve the spatiotemporal sequence problem.After analyzing the structure principle and characteristics of ConvGRU,the Encdoer-Decoder structure combined with ConvGRU is applied to the correction of meteorological elements.Finally,according to the experiment of the correction problem of meteorological elements in ConvGRU structure,a MeLC-GRU model based on spatiotemporal sequence is proposed,and the MeLC-GRU model is further optimized during the experiment.In this paper,the EC live data is used as the training target.The MeLC-GRU model is trained by using the historical EC live data combined with the current EC model forecast data.Finally,the trained MeLC-GRU model is corrected and evaluated.The experimental results show that compared with the numerical forecasting products and other revised algorithms,the MeLC-GRU model-based meteorological element grid correction algorithm has better correction effect on numerical forecasting and good spatio-temporal feature extraction performance.And it shows the certain advantages of model intelligent automation in the high-resolution forecasting.