| Since the clustering identification method is a key link in lightning early warning and forecasting technology,the accuracy of it is very important.There are many clustering methods in the existing research work,but different clustering methods have different effects in algorithm.There are obvious differences,and the thunderstorm identification effect may have a certain difference.Therefore,these clustering methods are significantly different in nature,and it is necessary to clarify the differences and effects of different clustering methods,and also provide a necessary reference for the selection of relatively reasonable clustering methods for different thunderstorm types.However,at this stage,there are few reports on the comprehensive comparison of different cluster analysis effects in thunderstorm identification,tracking and extrapolation.Therefore,the reason why this article uses three clustering methods is to evaluate the most stable and applicable method through the clustering effects and extrapolation effects of the three methods under different thunderstorm stages and different lightning distributions.This paper intends to establish a lightning approach warning system based on three clustering recognition algorithms as the core,so that users can more intuitively clarify the difference.The main results of this paper are as follows:(1)This paper uses the national ground lightning positioning data and radar reflectivity data,and uses DBSCAN,CFSFDP and E＿CFSFDP three clustering algorithms to analyze the27o—30oN,114o—117oE area from 11:00 to 18:00 Beijing time on September 21,2018.The four moments of a thunderstorm process were clustered and identified,and the advantages and disadvantages of various clustering algorithms were discussed by region and time period.The accuracy of DBSCAN algorithm is closely related to the distribution of data.The experimental results show that:(1)DBSCAN algorithm is only suitable for dealing with the situation where the density distribution of different clusters is relatively balanced;(2)CFSFDP algorithm does not consider the data spatial distribution characteristics due to the specified global density threshold d_c,the local flash data density and class When the spacing is unevenly distributed as shown in Figure 4-5(b)or when there is a"no density peak"distribution,it will lead to a decrease in the clustering quality;The small scale is further subdivided,and the metric function is used to selectively fuse,which solves the problem that the recognition rate of the CFSFDP algorithm is often not high in the thunderstorm cloud fusion stage.(2)Based on the national ground lightning location data and thunderstorm data identified by clustering,this paper uses the Kalman filter algorithm to extrapolate the thunderstorms in the area of 28o-32oN,108o-112oE from 8:00 to 9:00 Beijing time.The experimental results show that:(1)There is a correlation between the accuracy of thunderstorm extrapolation and the recognition accuracy,and the higher the recognition accuracy,the better the extrapolation effect.Therefore,the extrapolation results of thunderstorm identification based on E＿CFSFDP clustering algorithm are better than those based on CFSFDP and E＿CFSFDP.(2)Judging from the results of extrapolation and verification,the accuracy of the thunderstorm extrapolation algorithm based on Kalman filter presents a monotonous downward trend over time.The average extrapolation accuracy rate of the algorithm is about 50%within 60 minutes,and the extrapolation accuracy rate based on the three recognition algorithms is mostly less than 50%from 30 minutes onwards.(3)This paper establishes a lightning approach warning system based on the Py Qt framework,which realizes the functions of data storage,data management,thunderstorm identification and extrapolation,and provides users with more intuitive extrapolation and comparison results. |
PDF Full Text Request