Study in Ann Leak-Detection Method for Natural Gas Pipeline Based on Acoustic Signal

摘要：介绍了输气管道安全检测中以音波信号特征值为泄漏工况判断依据，以人工神经网络为泄漏工况判断标准的泄漏检测方法。首先，对用多种信号处理方法得到的音波信号特征值进行优选。其次，应用不同的人工神经网络对优选出的特征值进行处理，通过实际案例检验网络模态识别的能力，并针对每一个网络的处理效果（网络复杂度，训练时间和模态识别精确度）进行对比分析，从而优选出训练时间短，识别精确度高同时具有多模态识别能力的网络模型。最后将优选出的神经网络与实时音波信号采集系统相结合，编制出输气管道音波法泄漏检测系统软件。

关键词：泄漏检测 信号处理 人工神经网络 软件编制

Abstract  Introducing a leak–detection method of natural gas pipeline based on artificial neural network and acoustic signal. Taking the given acoustic characteristics of gas pipeline as different network’s input , comparing the efficiencies and results (including network’s complexity, training time and accuracy of their recognition system )of these networks, optimizing the parameters (training method ,epoch , performance , network’s construct and gradient) and finding out the most appropriate network for leak-finder. Finally, we combine the selected network and characteristics to program the leak detection system of natural gas pipeline.

Keyword  leak detection  signal processing  artificial neural network  program