Leakage Detection Of Natural Gas Pipeline Using RPROP-MDE Optimized Fuzzy Neural Network

For natural gas pipelines,it is critical to determine whether the pipeline has leak or not.Most of the conventional pipeline leak detection methods are exist the problem of locate difficulty and low detection accuracy,so it is difficult to apply to the natural gas pipeline.For this reason,using the fuzzy neural network to analyze the collection natural gas pipeline acoustic signals,in order to analyze the state of the pipeline leak and provide the basis for the leak judgment.Through the research for the actal situation of natural gas pipeline and the analysis for the actual leak data,also reference relevant literature and theoretical,base on these the natural gas pipeline leak detection method which based on RPROP-MDE optimized fuzzy RBF neural network is proposed.Base on natural gas pipelines acoustic signals characteristic,the signal denosing pretreatment is using wavelet transform method.And simulation conclusion is get that using soft threshold function and five layer dbl wavelet basis decomposition to de-noising for the acquisition of natural gas pipeline acoustic signal,the effect is more ideal,finally using wavelet decomposition to extracted the time domain and frequency domain characteristic value as the fuzzy RBF neural network input parameters.Base on the natural gas pipeline features the fuzzy RBF neural network leak detection model is built.And using the extract cancellation eigenvalues as the fuzzy RBF neural network input parameters,then through the fuzzy neural network to determine the pipelineleak situation.Against the defects of Differential evolution algorithm,from the three aspects of the crossover probability,variation factor,mutation strategy and cyclic variation of population location to improve the standard differential algorithm and using the typical Bencmark function to prove the improved differential algorithm has greatly improved in convergence performance and the obtained optimal solution.To avoid the influence of initial parameters to network training and the defects of Negative gradient algorithm in the adjustment of the network weights.The method which using RPROP-MDE joint algorithm to train fuzzy RBF network is proposed.That is using improved differential algorithm to optimized the Fuzzy Neural Network initial weights,then using RPROP algorithm to adjust the value of weight to further improve the network’s the prediction accuracy and convergence capability,finally through the actual simulation results to prove the improved algoritm is effectiveness.