Research On Lightning Strength Grade And Near Trend Forecast Model Based On Clustering

The quantification of strength grade of lightning and its spatial distribution are the improvement of the research area of the quantification of lightning strength.And the link between the two analyzed with land use type has the both theoretical and practical significance to the lightning protection.The trend forecast of the arrival of lighting is always a hotspot of lightning forecast.The clustering idea coming from the data mining is used in the related fields of lightning research to divide lightning strength,and the relationship between the spatial distribution of each strength grade and the corresponding land use types is discussed.At the same time,according to the shortcomings of the lightning forecast model built based on the method of DBSCAN proposed by predecessors,the algorithm of PDBSCAN is put forward by combining two optimization algorithms based on the method of DBSCAN.An instance of thunderbolt weather which occurs within Huaian in July 6,2009 is predicted and verified by using the algorithm of PDBSCAN.The experimental results show that the algorithm of PDBSCAN proposed in this paper to predict lightning trends is scientific and effective.The results are as follows:(1)according to the own characteristics of lightning data,K-Means clustering method is used to cluster strength grade of summer half year when occurrence rate of lightning is high in Jiangsu province from 2007 to 2009 in the environment of R.And five lightning clusters are obtained.The centroids of each lightning cluster of three years are respectively 19kA,31 kA,47 kA,75 kA and 138 kA.The number in the first and second estate accounted for more than 79.64%of the lightning intensity level is the biggest.The number in the later three levels is a decreasing trend.Huaian city belongs to the first estate by calculating the average of each level after clustering lightning intensity in Jiangsu province.In this city,the change of lightning intensity is the biggest and the situationis most complicate.Yancheng and Nanjing belong to the second estate.Xuzhou,Yangzhou and Suzhou belong to the third estate.Suqian,Taizhou,Nantong,Zhenjiang,Changzhou and Wuxi belong to the fourth estate.Only Lianyungang belongs to the fifth estate.(2)Land use types in Jiangsu province are inverted by using remote sensing data.And different types of land use are matched to corresponding lightning intensity level.It is found that woodland is the most closely related to each lightning intensity grade on unit area.The corresponding densities of the other five land use types are respectively 1.37,1.23,1.03,0.40 and 0.02(d/km2).In addition,there is an inflection point which reaches the peak in the second estate of lightning strength where center lightning strength is 31kA.And then begin to decay.The peak values are 0.88,1.13,1.05,1.14,0.93 and 1.23(d/km2).(3)Through studying the previous studies,the algorithm of PDBSCAN is put forward by combining two optimization algorithms.The algorithm assigns values to threshold MinPts and Domain radius by using the method of k-dist diagram and DK diagram.Starting from the data itself,updating the above two parameters in real time replaces empirical valuation method.At the same time,it also improved the scope of application of lightning cluster.The storage structure of the data set is replaced by the temporary connection table which plays the role of releasing the memory space and improving the operation efficiency.The center coordinate of maximum density cluster space of the lightning of the next period of time can be predicted by linear fitting centers of maximum cloud-to-ground discharge cluster in each period.It is found that the difference between the prediction result and the actual result is about 3.114km by the simulation of the instance of lightning weather.And it is proved that the model is effective to predict the moving trend of the core of the lightning.