Research On Pipeline Leakage Technology Based On CEEMDAN-RCMDE-LSTM

In this paper,taking the pipeline pressure wave signal as the research object,aiming at the problems of excessive noise influence,difficulty in extracting leakage characteristics,and low working condition identification accuracy of long-distance oil and gas pipelines,the pressure signals under different working conditions are collected by using laboratory pipelines,and the complete set of empirical mode decomposition with adaptive white noise is used to de-denoise the pipeline signals by adding adaptive white noise and the improved wavelet threshold denoising fusion algorithm.For the pressure signal after noise reduction,the fine composite multi-scale scatter entropy is extracted as the feature vector of the extraction,and finally the feature sequence is used to train the long and short time neural network model to identify and classify different working conditions of the pipeline.Firstly,the empirical modal decomposition algorithm and the CEEMDAN algorithm theory improved on this basis are studied in depth,and the relevant knowledge of improving the wavelet threshold denoising algorithm is studied.Using the two algorithms combined with the processing of non-stationary signals Heavy sine,Doppler curves,compared with EMD,CEEMDAN,wavelet threshold denoising and other algorithms,the algorithm robustness and noise reduction ability are analyzed,and the signal-to-noise ratio obtained by processing the signal by the fusion algorithm is found to be the highest,and the feasibility of the fusion algorithm processing pipeline pressure signal is initially verified.Aiming at the characteristic information of the significant component in the enhanced pipeline pressure signal and the problem that the modality is difficult to select after CEEMDAN decomposition,the Hausdorff distance is introduced,the pipeline signal is first decomposed by CEEMDAN,and the Significant component,abnormal component and useless component are screened by HD,and the abnormal component and significant component processed by the improved wavelet threshold denoising algorithm are reconstructed to obtain the noise reduction signal.By evaluating the processing effect by correlation coefficient and root mean square error,it is found that the fusion denoising algorithm can effectively remove noise and effectively retain leakage characteristics.Secondly,aiming at the problem that the characteristics of the tiny leakage signal of the pipeline are difficult to extract,the pipeline feature extraction method based on information entropy is proposed,and the fine composite multi-scale dispersion entropy is used to solve the problem that the single-scale dispersion entropy cannot fully reflect all the feature information leaked,and the problem that the multi-scale dispersion entropy decreases with the increase of the scale factor.Finally,taking the identification of natural gas pipeline conditions as the starting point,a variety of machine learning algorithms are used to process the characteristics of pipeline pressure signals,a pipeline leak detection model is trained,and a pipeline leak detection system based on long-term and short-term neural networks is proposed,and the recognition accuracy and recognition rate of the analysis algorithm are compared.Experiments show that the use of LSTM neural network for condition recognition can further improve the detection accuracy and efficiency of the model.