Research On Feature Extraction Method Of Multi-mode Acoustic Signal Of Natural Gas Pipeline Based On Variational Mode Decomposition

Natural gas,as a low-carbon fossil energy,has the advantages of being clean,stable,and efficient,and is the main energy source supporting the country’s “dual carbon”goals.Pipelines,due to their convenience,efficiency,and economy,are the ideal carrier for constructing long-distance natural gas transportation systems.However,with the length of pipeline construction and the age of the pipeline increasing,pipeline leakage events will inevitably occur.When a natural gas pipeline leaks,the gas sprays outward and generates a sound wave signal.Due to the complexity of the pipeline’s service environment and the variability of actual working conditions,the sound wave signals collected by sensors exhibit nonlinearity and non-stationarity,and the signal characteristics are not easy to extract.This can lead to false alarms and missed reports in pipeline leakage detection.Therefore,this dissertation is based on the Variational Mode Decomposition(VMD)algorithm and conducts in-depth and systematic research on the feature extraction technology of pipeline sound wave signals under different working conditions.Its purpose is to explore new methods for preprocessing and feature extraction of natural gas pipeline sound wave signals,to solve the problem of difficult feature extraction of natural gas pipeline leakage sound wave signals,and to improve the accuracy of leakage detection,thereby laying a solid theoretical foundation for the establishment of intelligent natural gas pipeline health status monitoring and leakage detection technology.This dissertation mainly completes the following research content:(1)In response to the problem that the noise contained in the sound wave signals collected by the sound wave sensor during pipeline leakage will affect the feature extraction of the signal and the accuracy of leakage detection,a VMD-WD denoising method is proposed.This method first uses the Minimum Bayesian Distance method to determine the decomposition layer K value of VMD,and then selects effective mode components by evaluating the Wasserstein Distance(WD)values between the probability density of each Intrinsic Mode Function(IMF)component decomposed by VMD and the original signal.Finally,the selected effective mode components are reconstructed to achieve signal denoising.Experiments and analysis on both simulated signals and pipeline signals were conducted to verify the feasibility and superiority of the proposed method through algorithm testing and performance analysis.The method was also applied to the denoising processing of pipeline signals.(2)To address the issue that the characteristics of leakage signals are easily overwhelmed by strong noise in the environment,a pipeline signal single feature extraction method based on VMD denoising and Dispersion Entropy(DE)combination is proposed.This method first uses the VMD-WD method to denoise the collected signal.Then,the DE value of the denoised signal is calculated as the feature parameter of the signal,thereby achieving feature extraction of the signal.Through experimental analysis,this method reduces the interference of noise on signal features,and combines DE parameters to achieve extraction of normal signal,knocking signal,and leakage signal features in strong noise environments.The extreme learning machine(ELM)is used to identify the characteristics of the three working conditions signals,and comparative experiments are conducted to verify the relative superiority of the proposed feature extraction method.(3)A method for feature extraction of natural gas pipeline acoustic signals based on VMD denoising and multiple feature fusion is proposed based on the idea of feature fusion to address the difficulty in extracting leakage signal features of different apertures in natural gas pipelines.The method first uses the VMD-WD method to denoise the collected signal,and then selects multiple feature parameters from different angles to analyze the signal.By analyzing and selecting multiple feature parameters that can effectively distinguish different working condition signals,the values of multiple feature parameters of the denoised signal are calculated.Finally,the feature parameter values are combined to form a feature vector to achieve feature extraction of the signal.Through experimental analysis,the method effectively extracts the features of leakage signals of different apertures.The ELM is used to identify the extracted signal features,and through comparative experiments,the results confirm that the method’s feature extraction is superior to the feature extraction method based on single feature parameters.(4)To cope with the difficulty of extracting signal features due to the similarity of waveform characteristics between single-point and two-point leakage working conditions,this dissertation proposes a natural gas pipeline acoustic signal feature extraction method based on the combination of feature components and Exponential Entropy(EE)by drawing on the idea of local features in signals.The method first uses the VMD algorithm to decompose the signal into multiple IMF components.Then,by analyzing the correlation between the IMF components and the original signal,the IMF component with a high correlation with the original signal is selected as the feature component.Finally,the EE value of the feature component is calculated as the signal’s feature parameter to form a feature vector,achieving the extraction of signal features.Through experimental analysis of single-point leakage signals,two-point leakage signals,and normal signals,this method can not only overcome the interference of noise components on signal feature extraction by selecting feature components but also better extract the feature information of the original signal by analyzing local feature components.The ELM is used to identify the features of the three working condition signals extracted in this study.Comparative analysis with other methods confirms the feasibility and superiority of the proposed feature extraction method.(5)Due to the complexity of the acoustic signals in different working conditions of natural gas pipelines,the dynamic characteristics on a single scale cannot fully express all signal feature information.To address this,a comprehensive indicator,Partial Mean of Multi-scale Dispersion Entropy(MDEPM)based on multi-scale dispersion entropy and partial mean,is proposed as a feature parameter for natural gas pipeline acoustic signals.Then,a new feature extraction method,IMF-KSD-MDEPM,is proposed based on the local feature information of MDEPM values.In this method,Kolmogorov-Smirnov Distance(KSD)is used as the indicator for selecting feature components,and the KSD values between each IMF component after VMD decomposition and the original signal are calculated.Feature components with smaller KSD values are selected.Finally,the MDEPM values of the feature components are calculated to form a feature vector,comprehensively reflecting the multi-scale characteristics of signals in different working conditions.Through experimental analysis of various working condition signals in natural gas pipelines,this method can extract features of signals in different conditions and is more conducive to the classification and identification of different working condition signals by the classifier.