Improving The Quality Of Model For Severe Convection Weather Classification And Recognition

Severe convection weather is one of the main causes of meteorological disasters,which mainly includes hail,tornado and thunderstorm wind.In China,the strong convective weather caused serious casualties and property losses every year.Under the requirement of meteorological research,our research team has developed a classification and identification system for severe convection weather,which takes doppler weather radar data as the main input to analyze,identify and forecast severe convection weather,and output text files containing feature information of severe convection weather and cell.Based on the existing classification and recognition model,this paper improves and expands the function of it.The main work is as follows:(1)In this paper,the databases of monomer and its forecast information,as well as the actual situation of wind speed,precipitation and hail are established.At the same time,a visual interactive interface is established through QT to display forecast and actual information such as monomer,gale,precipitation and hail on the map.Then the monomer timing sequence is analyzed to give the monomer timing sequence characteristic trend graph,and meteorological evaluation is carried out based on the actual situation and forecast information,and the hit rate,null rate and missing rate of the forecast result are calculated and written into the evaluation database.(2)Due to the fact that the hourly precipitation R and the radar reflectivity factor value Z measured by the automatic station in the case of hail fall did not meet the similar power index relationship,the abnormal phenomenon of high reflectivity value and low precipitation occurred.In this paper,a large number of actual hail data and automatic station precipitation data corresponding to its location are collected to screen out the automatic station data with abnormal precipitation and give the abnormal score.According to this score,hail clouds could be judged and used to assist the marking of hail samples.(3)On the basis of the existing classification model of hail precipitation,this paper used the collection of hail and precipitation sample data to build a Wayne predictor,which was used to give the probability of hail or heavy precipitation in the current cell.(4)Based on the improvement of the deep learning model for classification and recognition of strong convection,the data of hail precipitation model is enhanced,the network structure of the model before optimization is introduced by residual network,and the hail precipitation model is superparametric optimized;the sample data structure of the wind model is modified,and the double input network structure model is proposed,and the gale model is optimized.