Research On Feature Extraction And Detection Technology Of Natural Gas Pipeline Leakage Based On Variational Mode Decomposition

In view of the difficulty of extracting the leakage characteristic information in natural gas pipeline leakage,variational mode decomposition(VMD)is applied in pipeline leak detection to realize original signal denoising,and then characteristic value are extracted by the feature entropy of the cloud model.Finally,the main parameters of Support Vector Machines(SVM)are optimized using the grid search method,thereby improving the accuracy and accuracy of the condition recognition.Firstly,the theoretical algorithm of VMD is deeply studied.Through the analysis of its noise robustness,separation of low-frequency signal,and the processing capability for non-stationary signals,it is proved the superiority and effectiveness of VMD compared with empirical mode decomposition(EMD)and wavelet decomposition.Secondly,aiming at the problem that the characteristic information of the pipeline leakage signal is confused during the transmission process,a denoising algorithm based on VMD is proposed.To solve the problem of selecting effective components of VMD,a joint denoising algorithm based on VMD and Hausdorff distance(HD)is proposed.This method decomposes the original signal into multiple band-limited intrinsic mode function(BLIMF)components,and then uses HD to measure the similarity between the original signal and each BLIMF,and selects relevant components to reconstruct signal.The denoising effect is evaluated by detrended fluctuation analysis(DFA),and ? =1.8992,which means that VMD-HD can effectively remove noise and obtain smoother filtering signals.Thirdly,aiming at the non-stationarity and uncertainty of natural gas pipeline leakage signal,a feature extraction method based on VMD and cloud model characteristic entropy is proposed.And the selection of the default scale K value in VMD decomposition process is also studied.The method uses VMD algorithm to decompose the signal to obtain BLIMFs,and selects the main mode components containing a large amount of leakage characteristic information by the correlation coefficient,and then calculates the cloud model characteristic entropy of main modes as the characteristic parameters of signal by backward cloud generator.Finally centroid frequency(FC)is also input as characteristic parameters into the SVM.Results show that the method is feasible for pipeline fault diagnosis.Finally,aiming at the problem that the finite number of samples in this paper,which affects the classification accuracy and generalization ability,the support vector machine(SVM)algorithm is used in this paper to classify the different pipeline condition signals.The libsvm toolbox was used to study the ability of different kernel functions to classify the three types of pipeline signals(normal signal,leakage signal,and interference signal).The grid search method is used to optimize the c and g parameters,and a classification accuracy of 98.33% is obtained,which can effectively distinguish different signals.