The Study Of Pipeline Leakage Signal Extraction Based On Underdetermined Blind Source Separation

As a bridge to connect the long-distance pipeline and the end-user, city high and intermediate pressure gas pipeline is rapidly increasing with city scale expansion. However, once the pipeline leakage happens, it will cause serious environmental pollution,even danger to life safety. Most of the Pipeline leakage detection methods at present mainly research the pipeline leakage detection and location using signal processing method. However, the premise of this method is the leakage signal has been acquired, actually, effective leakage signals are often mixed with a variety of interference signals, thus the premise of the leakage detection and location is obtaining the effective leakage signal accurately. In this paper, to achieve accurate detection of pipeline leakage information, based on underdetermined blind source separation are studied, the effective separation to the collected observation signal with the underdetermined blind source separation method and the reconstruction of the source signal based on compressed sensing theory. This paper will focus on the following works:Firstly, on the basis of analyzing the development and research statuses of detection technology method at home and abroad in pipeline leakage, the paper introduces the development situation of blind source separation and the application of the blind source separation in pipeline leakage signal extraction.Then, a kind of mixed matrix estimation method based on signal linear aggregation degree is studied and put forward. This method first makes the short-time Fourier transform to the observation signal collected by sensor, and second, measures signal linear aggregation degree in the time-frequency domain and enhances signal linear aggregation degree with the monophyletic geometric factor, and at last, estimates the mixing matrix using k-means clustering algorithm in optimized initial clustering center.Furthermore, according to the consistency of the underdetermined blind source separation and the compression model under compressed sensing, the underdetermined blind source separation problem is transformed into the reconfiguration problem of sparse signal in compression perception theory. An over-complete dictionary training method is adopted to realize signal sparse representation and access to dictionaries and the orthogonal matching pursuit is used to realize the source signal reconstruction and finally to realize the separation of each source signal.Finally, through experiments the method this thesis proposes that extracting pipeline leakage signal based on the underdetermined blind source separation is verified, which reduces the requirement for signal sparsity and could accurately estimate the mixing matrix and source signal. The experimental result shows that this method has feasibility and effectiveness.