Application Research Of VMD And ELM Algorithms In Natural Gas Pipeline Leak Detection

The infrasound signal generated by the leakage of natural gas pipeline is a non-stationary signal,There are some complicated problems in the actual gas pipeline leakage detection process.For example,the extraction of leakage feature information is difficult and often results in misstatements,false positives,and false negatives.To solve the above problems,Variational Mode Decomposition(VMD)algorithm is applied in pipeline leak detection to complete the pipeline leakage signal decomposition,denoising,feature extraction and classification recognition.First of all,the theoretical knowledge of VMD is studied and applied to noise robustness,low-frequency signal separation and real pipeline leakage signal processing.Comparing the results of the de-noising and de-noising simulation experiments of pipeline signal with Empirical Mode Decomposition(EMD)and Ensemble Empirical Mode Decomposition,it is proved that the VMD algorithm is feasible in the detection of natural gas pipeline leakage sex and decomposition denoising effectiveness.Secondly,introducing the entropy theory for measuring the complexity of signals.This chapter focuses on the theory of sample entropy,and based on the sample entropy theory,a feature extraction method based on VMD and sample entropy is proposed.When using VMD to decompose pipeline signals,in order to prevent over-decomposition or under-decomposition problems,the determination of the preset scale K through the center frequency is studied,so that the signal is optimally decomposed.Then,the signal is decomposed into several modal function components IMFs by using the VMD algorithm.Then a modal component containing a large amount of feature information of the original signal is selected as an effective IMF component using a correlation coefficient criterion,and the sample entropy of these selected effective components is finally calculated.,And sample entropy normalization process to get the normalized feature vector.Prepare for the classification and identification of subsequent pipeline condition signals.Finally,taking the sonic signal data as the research object which collected from the gas pipeline leak detection laboratory,the signals of three types of pipeline conditions are studied.Extreme learning machine(ELM)algorithm is used to classify and recognize the signals of three types of pipeline conditions.The results show that ELM has a good recognition effect on different pipeline conditions,and the recognition accuracy is as high as 91.11%.