Research On Leak Detection And Location Of Gas Pipeline Based On Stacked AutoEncoders

At present,the construction of natural gas pipeline is developing rapidly,but the frequent leakage accidents caused by corrosion,aging,man-made damage and other factors of pipeline bring many adverse effects on daily production and life.In order to accurately locate the pipeline leakage and identify the leakage caliber,the intelligent fault diagnosis of the pipeline leakage signal,the identification of the pipeline leakage caliber and the detection of the leakage localization are realized in this paper based on the deep learning theory and driven by big data.At the same time,combined with the relevant tests,the effectiveness of the method is verified.In terms of fault diagnosis,combining the advantages of big data and deep learning,the stacked Autoncoders(SAE)is used to build deep neural networks(DNN)in this study.The input end is the signal spectrum of pipeline leakage,which realizes the automatic extraction and classification of fault features,avoiding the defects of traditional methods in feature extraction and data classification.At the same time,the layer by layer learning characteristics of SAE are explored preliminarily.Through data visualization,the feature learning process of SAE is represented layer by layer.The results show that with the increase of the number of learning network layers,the leakage characteristic data of the pipeline is gradually clear and obvious,and the leakage signal characteristics of different caliber and distance show obviously different trends.In the aspect of leakage location,aiming at the problems of large amount of data,long training time and too many iterations of neural network,in the process of pipeline leakage detection and intelligent fault diagnosis,the batch normalization(BN)algorithm is introduced to make neural network realize fast and intelligent fault diagnosis.In this study,any activation layer of SAE is batch normalization to make any single layer network input have a stable distribution and improve the training effect of the network.The experimental results show that the SAE method with batch normalization algorithm can improve the effect of fault location when the training samples are small,the training time and the number of iterations is small.In the face of the problem that the diagnosis effect is not good under the condition of insufficient samples,this paper proposes Bootstrap-SAE,which expands the number of samples by re self-help sampling of data,and simulates the real leakage signal to generate artificial signals,so as to increase the number of samples generated by different positioning distances.Then feature extraction is carried out to achieve the accurate classification of different fault types.The leakage test with insufficient samples shows that the method can significantly improve the accuracy in the case of insufficient samples.The experiment shows that the intelligent diagnosis based on SAE has a good effect in pipeline leakage detection and location.It has certain advantages for the identification and location of leakage caliber,and provides theoretical support and application prospect for practical engineering application.