| With the continuous advancement of industrialization in China,the safe and stable operation of transmission and distribution lines has become the basic guarantee of continuous power supply for residents.However,transmission and distribution line faults caused by aging and environmental factors are inevitable.The research shows that 40%-70% of the fault of transmission and distribution lines is caused by lightning strike.Improving the accuracy of thunder recognition method is the research hotspot of power system at present.However,the operation environment of transmission and distribution lines is relatively complex,most of the lines are bad.In the middle,it needs to pass through the plateau,plain and other field areas or the combination of urban and rural areas,such as commercial areas,industrial areas,residential areas and so on.In this complex and changeable environment,the thunder signal collected is accompanied by a variety of interference signals.In this paper,based on this background,the thunder recognition technology in complex environment is studied.Firstly,aiming at the phenomenon of thermal noise in thunder signal collected by audio acquisition equipment,one-dimensional wavelet denoising and wavelet packet denoising methods are used to denoise the audio signal,and the selection of the optimal parameters in the process of denoising is analyzed.The denoising quality of the two methods is evaluated by the correlation of signal-to-noise ratio and time-domain diagram.In the processing of thunder noise reduction,the one-dimensional wavelet denoising method is better.Then,the spectrum analysis of several sections of thunder and interference audio is carried out,and the main frequency bands of the short-term energy distribution in the thunder sound frequency are determined.The signal in the frequency band is filtered by low-pass filtering method,and the effective signal in the frequency band of thunder sound energy distribution is retained.The singular values of linear prediction cepstrum coefficient,Mel cepstrum coefficient and empirical mode decomposition are extracted respectively for the filtered signal.The feature vectors are constructed based on information entropy,and the multi-dimensional feature vectors are reduced to preserve the effective feature information.Finally,a support vector machine classifier based on radial basis function(RBF)is constructed,and the classifier is used to train and recognize the feature vectors of audio signals.The experimental results show that the recognition effect of extracting singular value and information entropy as eigenvectors by empirical mode decomposition is the best,and the recognition rate can reach 99.26%.The experimental results show that it has good stability for the research of thunder recognition method in complex environment. |
