压缩感知结合卷积网络的天然气管道泄漏孔径识别

Recognition of leakage aperture of natural gas pipeline based on compression sensing and convolution network

针对传统天然气管道泄漏孔径检测面临的原始数据冗余性大、特征选取主观依赖性强以及复杂环境下识别准确率低等问题,提出了一种将压缩感知与深度卷积神经网络相结合的泄漏孔径识别方法。首先利用随机高斯矩阵对原始泄漏信号进行压缩采集,以较少的压缩感知域数据获取全部泄漏信息;然后构建深度一维卷积神经网络,将压缩采集数据送入网络中实现自适应特征提取及高准确度泄漏孔径识别;还对主要参数的影响进行了深入的分析。实验结果表明,该方法能够快速、准确地实现天然气管道泄漏孔径识别,且在低信噪比环境下具有较好的鲁棒性,总体识别效果优于传统的分类方法。

Aiming at problems of large redundancy of raw data, strong subjectivity dependence of feature selection and low recognition accuracy under complex environment for traditional natural gas pipeline leakage aperture recognition, a leakage aperture recognition method based on compression sensing and 1-D convolution network was proposed. Firstly, the random Gaussian matrix was used to do compression collection of original leakage signals, and the full leakage information was obtained with less compression sensing domain data. Then, a deep 1-D convolutional network was constructed, and the compression collection data were fed into the network to realize adaptive feature extraction and leakage aperture recognition with high accuracy. Finally, effects of the main parameters on recognition results were analyzed. Test results showed that the proposed method can quickly and accurately realize the leakage aperture identification of natural gas pipelines;it has better robustness under low SNR environment;its overall recognition effect is superior to that of the traditional classification method.