Study On Gas Pipeline Leak Detection And Location Based On Wavelet Packet And Fuzzy Support Vector Machine

Pipeline plays a very important role in petroleum, chemical, natural gas, urban construction and other industries. With the development of world economy,the worldwide natural gas transport and distribution network becomes a more complex and continuously expanding one. In recent years, with the rapid development in the pipeline network, pipeline accidents occur frequently. Combining the experimental with theoretical approaches, the studies on detecting and locating leakages in straight and branch gas pipelines are done as following:1) Constructing of the experimental platform. A gas pipeline leakage detection test bench is built based on the fluid dynamics theory and the signal detection methods. With the test bench, such simulink experiments on straight and branch gas pipelines as online monitoring, inspection and location can be perfromed in different leakage aperture and different pressure conditions.2) According to the characteristics of the pipeline leakage acoustic emission signal, wavelet packet analysis method is adopted to extract signal features. To improve the spectrum mixing effects brought by the traditional wavelet packet when decomposing and reconstructing signal, this dissertation proposes an improved wavelet packet algorithm. The experimental results show the better effect of the improved wavelet packet analysis method in feature extraction. The improved algorithm can overcome the mixing effects and accurately extract the signal features which can reflect all kinds of information of different states, and the feature can comprehensive describe the characteristics of the pipeline leakage signal. In this dissertation, we adopt the improved wavelet packet algorithm for acoustic emission signal feature extraction.3) To improve the training efficiency of the fuzzy support vector machines(FSVM), this dissertation adopts the fuzzy kernel clustering algorithms combing with the support vector distribution rules to preprocess the training set. In which, the sample close to the center of the class and away from classification borders will not be the support vector, having no contribution to learning machine, and be removed from the training set through clustering analysis. The experiments on three kinds of data prove the feasibility of this method.4) To reduce the influence of noise samples and decrease the rate of error classification, the dissertation introduces the membership functions and fuzzy support vector machine(FSVM). The member ship function based on class center assigns smaller membership value to the support vector, which can reduce the classification accuracy. Based on this viewpoint, this dissertation introduces an improved method to design membership function, substituting hyper-plane in class for class center and defining membership function according to the distance between each type of data and the hyper-plane to increase the punishment for fallible sample points, while assigning small membership value to the sample far away from the classification hyper-plane and making the sample not support vector. The experiments on artificial datasets and standard data sets prove the efficiency of this algorithm and the better classification results.5) In order to reduce the error in each processing stage and improve the positioning accuracy, the paper adopts the improved method to calculate time delay which is used for positioning. The paper get the time delay based on correlation time delay estimation, the leakage and normal signal are decomposed and reconstructed by improved wavelet packet algorithm, and then get the single-subband signals, we compare the energy of the corresponding single-subband signals got from normal and leakage signal and choose two single-subband signal whose energy have the biggest change before and after leakage to calculate time delay, and then we construct weighting factor based on the energy of single-subband signal. Final, we obtain the final time delay by weighted multiple time delay; The final leakage coordinate is calculated through weighted multiple leakage coordinates got from different sensors combination. The experimental results show that the proposed method can improve positioning accuracy and is feasible.With the development of municipal gas, the operating safety problem of the pipeline becomes much more important. Study on the detection and location of municipal gas pipeline system can not only ensure the security of gas supply system and reduce accident damage, but also enhance the security management of municipal gas supply system, prolong the service time of the system, and improve its economic and social benefits.