Research Of The Natural Gas Pipeline Leakage Based On EEMD And SVM

With the ascension of natural gas’ strategic resources role and status, the security of the pipeline transport more and more gets people’s attention. During the pipeline running and maintenance, need to monitor the pipeline in real time, accurately judge the running status and timely diagnose the hidden trouble of pipeline leakage to avoid safety accident. Although pipeline leak detection technology continuously improves, but there are still falsely reported situations in pipeline leak detection. Therefore, in view of this situation, this thesis gives the design of the gas pipeline leakage diagnosis, to achieve highly accurate intelligent diagnosis.This thesis applies EMD empirical mode decomposition method for sound pipeline leakage signal processing. This method can be signals of different scales(frequency) of trend of fluctuations or broken down step by step, to form intrinsic mode functions. Extract approximate entropy and energy entropy features for the each intrinsic mode function reflecting the original signal characteristics. There is modal aliasing effect in the traditional empirical mode decomposition method. In depth analysis, this these eventually applies EEMD general empirical mode decomposition method with the approximate entropy and the energy entropy to extract feature. Apply support vector machine(SVM) classify for the feature vector groups extracted by four kinds of feature extraction method, contrast effect and determine the best way of feature extraction.Complexity and generalization ability of the support vector machine depend on the penalty factor C and the kernel function parameter g. In order to improve the accuracy of identification diagnoses, we require an accurate, quick and stable method to find the optimal parameters. This thesis respectively applies the grid search parameters optimization, particle swarm optimization algorithm, genetic algorithm, particle swarm combined with genetic algorithm to support vector machine classification’s penalty factor C and the kernel function parameter g parameter optimization with Libsvm software platform, and compares classification accuracy effect, finally achieves the goal of high pattern classification accuracy.