Study On The Implementation And Application Of Gas-pipeline's Leak-detection Based On Acoustic Signal

Nowadays,the industry of domestic oil & gas pipelines has been booming,and the pipeline's total length has reached to 100 thousand miles or more.However,this amazing expansion also brings more frequent accident of pipelines.So,we have to formulate an appropriate scheme or program a system of pipeline's leak detection and location to ease such accident's damage to public life and property or just prevent them.Currently,most of leak-detection systems which are in used take one-model(leak condition,normal condition)detection algorithm as system core-deciding program.Although,it is apply to the pipelines which have stable running-condition and little interfering conditions,but to those changeable running-condition(valve opening,compress shutting and so on)pipelines,that would be terrible.so,this article took pipeline's acoustic signals as input of leak detection,and developed an acoustic pipeline's leak detection and location system.First of all,this article introduces ANN into pipeline acoustic leak-detection method as a core-deciding algorithm to solve the problem of identification in pipelines' working conditions,Then,in order to get a best application effect,we prioritized a bunch of characteristic value in different signal's field(time domain,frequency domain and time-frequency domain),and combined the result of prioritizing select to analyze application effect of different ANN.At last,we come to the conclusion that BP network has higher accuracy rate of leak-detection and better ability to against interference than other networks.Secondly,BP neural network has a simple construction,available training algorithm,and it can infinitely approach the target in theory.At the same time,it also has some disadvantages,for example:Unstable of training convergence,falling into local optimum easily,strong sample-dependence and so on.So,we present some optimized algorithm aim toenhance BP network's generalization ability,stability of training convergence and increase accuracy rate of model-detection.In this article,we take Bias normalization algorithm,optimized adaptive genetic algorithm and fuzzy neural network to make up these three defect separately.After some testing trials,we can prove this optimized BP network has a better network performance and higher accuracy rate of model-detection.At the end,in order to program a pipeline leak detection and location system with high working efficiency and applicability,we choice Visual Basic 6.0 to program user interphase,because of its simple programing,abundant external interface and high hardware-applicability,and we took toolbox in MATLAB to write the core-detection algorithm,after all,we combined Microsoft Access in VB to record and display the alarming message of pipeline's leak.So,this article can be concluded as programing a pipeline leak detection and location system with acoustic sensor device,MATLAB,Visual Basic 6.0 and Microsoft Access,based on multi model-detection BP neural network,and acquiring a satisfactory result in laboratory.