Research On Pipeline Leak Detection Based On Local Characteristic-Scale Decomposition And Semi-Supervised FCM

Pipeline leak not only pollutes the environment,affects production,but also causes casualties,so the detection of pipeline leak is particularly important.Negative pressure wave method is the most widely used in pipeline leak detection.However,due to the interference of strong background noises and operating conditions,the method has less denoising effect,higher false alarm rate and lower leak location accuracy.Aiming at the above problems,this paper will combine local characteristic-scale decomposition(LCD),semi-supervised fuzzy C-means clustering(FCM)and second correlation method to improve the accuracy of pipeline leak detection.Non-local means(NLM)-empirical mode decomposition(EMD)autocorrelation denoising algorithm and the noise statistical LCD autocorrelation denoising algorithm are studied.In NLM-EMD autocorrelation algorithm,NLM is used as the pretreatment of EMD to improve the decomposition quality and denoising quality of EMD.In the noise statistical LCD autocorrelation algorithm,LCD is used to improve the endpoint effect and false component of EMD and denoise combined with noise statistics.Compared with common denoising algorithms under different signal to noise ratio(SNR),the results show that the noise statistical LCD autocorrelation algorithm can not only well denoise,but also perform well under low SNR.The leak detection algorithm based on semi-supervised FCM and Euclidean proximity method is studied.The energy entropy and trend term variance of normal signals,leak signals and valve signals after denoising are extracted to compose the feature matrix which is used to train semi-supervised FCM,and Euclidean proximity method is used to identify the leak signals.Compared this algorithm with unsupervised FCM and supervised PNN probabilistic neural network,the results show that semisupervised FCM has less markup workload and higher leak detection accuracy.FCM-second correlation adaptive delay estimation algorithm is studied to improve leak location accuracy.For the detected simulated leak signals,FCM is used to extract pressure drop signals with obvious leak characteristic to improve the performance of second correlation.Compared with common algorithms,the results show that the algorithm can locate pipeline leak signals with higher precision and its accuracy is obviously higher than second correlation method.The laboratory and field pipeline pressure signals are used to verify the leak detection effect of the algorithms of this paper.Pressure signals of different leak locations collected in the snake pipeline leak detection system of the laboratory are processed by the algorithms of this paper.At the same time,pressure signals at different leak locations in different lengths of oil pipelines in Xinjiang are processed,and the average relative location error is only 0.67%,which is much smaller than other common algorithms.It is proved that the algorithms of this paper can be effectively used in the research of oil pipeline leak detection.