Lightning Magnetic Field Spectrum Characteristics And Lightning Signal Classification

The 25 wavelets are used to decompose the magnetic field signals of the lightning detection network in Nanjing.It is found that when decomposing the cloud flash and CG(cloud-ground)signals,the db,sym,and gaus high-order mother wavelets showed oscillations at high frequencies,and the total energy trend of first order mother wavelets is the same.This paper uses a positive CG lightning at 2pm on July 26,2018 and a negative CG(cloud-ground)lightning at 2pm on August 3,2018 detected by Nanjing VLF/LF three-dimensional lightning detection network.A detailed comparative analysis of the spectral characteristics of different lightning discharge stages is carried out.The preliminary conclusions are as follows: 1)It can be seen from the threedimensional channel of these two lightning events that the two lightning events both have three return strokes.Moreover,from the positions of the radiation sources,it is obvious that these three return strokes develop along the same channel to the ground;2)The spectrum of the intra-cloud discharge process of both positive CG lightning and negative CG lightning is similar,and it is mainly concentrated above 40 k Hz;3)The spectrum of dart leader for both positive CG and negative CG lightning is concentrated above 100 k Hz.There is almost no difference between the spectrum of positive and negative leader,but the radiation frequency of leader process is much higher than that of the intra-cloud discharge process.4)The spectrum of positive and negative return stroke is similar,and it is mainly concentrated below 20 k Hz.Additionally,the high frequency component of the subsequent return stroke is less than that of the first return stroke.A lightning classification scheme based on wavelet is proposed.The accuracy for the 300 data is shown as follows: the recognition rate of in-cloud activities is 58%,the recognition rate of return stroke is 82%,and the recognition rate of steps is 84%.Onedimensional and two-dimensional CNN(convolutional neural network)lightning signal classification models are established.Using the normalized data,the one-dimensional signal recognition rate is 97.98%,and the two-dimensional signal recognition rate is 98.33%.It is much higher than the classification model based on wavelet.A onedimensional CNN lightning distance judgment model is established,then the 6 types of lightning data of 50 to 300 km are used to train the model.The accuracy of 50 km is the highest.As the distance increases,the accuracy decreases gradually.The training and verification accuracy using the original data is 37.37%,and the accuracy of the normalized data is 34%.