Research On Pipeline Leak Detection Method Based On VMD And Entropy Features

With the increasing demand for natural gas energy,the natural gas pipeline transportation industry has been developing rapidly.Affected by factors such as aging pipeline networks and human damage,pipeline leakage accidents occur frequently,which seriously threaten people’s lives and property safety.The early leakage of underground natural gas pipelines in most cities in China is small and often difficult to detect and repair,therefore,accurate and timely leak detection of pipelines can effectively avoid safety accidents.This paper mainly uses three key techniques of pre-processing,feature extraction and working condition identification to realize the detection of pipeline leakage signals.The main research contents are as follows.First,the principles of adaptive modal decomposition methods EMD,EEMD,CEEMD,MEEMD and VMD are analyzed in detail.Before the VMD decomposition,the number of decomposed modes K needs to be set artificially,and the improper setting of K value will seriously affect the performance of VMD decomposition.To address this problem,the Euclidean distance-based VMD method is proposed,and the parameter K value is determined by evaluating the Euclidean distance between the signal of each mode component of VMD decomposition and the original signal.The above modal decomposition method is used to decompose the simulated signal to obtain several modal components,and the Euclidean distance of each modal is calculated to find the modal components that are most similar to the original signal for reconstruction to realize the signal filtering process.In order to maximize the useful information of the original signal,the two methods are applied to the processing of the actual pipeline leakage signal.Secondly,since the useful features are not easily extracted due to the small amplitude variation of natural gas pipeline micro-leakage signals,the entropy theory,which measures the complexity of signals,is introduced,and a multi-scale fuzzy entropy feature extraction algorithm is proposed for the problem that the entropy features of natural gas pipeline microleakage signals are difficult to be fully extracted at a single scale.The multi-scale fuzzy entropy is extracted from the reconstructed signals of pipelines after MEEMD and improved VMD filtering processes to construct feature vectors,respectively.The effectiveness and superiority of the feature extraction method based on multiscale fuzzy entropy in the field of pipeline leakage detection are verified by comparing with multiscale aligned entropy.Finally,the basic theory of ELM algorithm is studied,and the PSO algorithm is used to optimize the input weights and implicit layer bias parameters of ELM to solve the problem of poor generalization ability caused by the randomness of ELM parameter settings.The basic PSO algorithm still has room for improvement.In this paper,we add the selection,crossover and mutation operations of GA on the basis of the basic PSO algorithm and propose the genetic particle swarm hybrid optimization algorithm(GA-PSO),which improves the possible premature convergence problem of PSO algorithm and improves the global search ability of the algorithm,and through the comparative analysis of various classification models,we prove that the GA-PSO-ELM classification model converges quickly and with high convergence accuracy,and the overall recognition of different working conditions of pipelines reaches 98.67%,which has a high accuracy rate.