阀门内漏识别及内漏速率量化技术研究

阀门作为天然气管线的关键部件,若发生内漏会带来经济损失及生产安全隐患。因此,阀门内漏的有效诊断及内漏速率的准确量化具有重大意义。针对复杂背景噪声下内漏诊断效率不高的问题,以内漏信号和非泄漏噪声信号的功率谱密度图作为输入,构建了阀门内漏卷积神经网络(convolutional neural network, CNN)识别模型;针对物理理论及浅层网络模型在多工况阀门内漏数据上存在量化误差大的问题,构建了阀门内漏速率深度信念网络(deep belief network, DBN)量化回归模型,并与支持向量回归机、BP神经网络等模型进行了对比研究。研究结果表明:所构建模型的内漏识别准确率及内漏速率量化平均绝对百分比误差分别为99%和9.101 2,证实了所构建模型的高效性,为阀门内漏诊断与评价开拓了新的研究方向。

A study on valve internal leakage identification and leakage rate quantification

Valves are a vital component in natural gas pipelines. If internal leakage occurs, it will bring economic losses and potential production safety hazards. Therefore, the effective diagnosis of valve internal leakage and accurate quantification of internal leakage rate are of great significance. Aiming at the problem of low efficiency of internal leakage diagnosis under complex background noise, based on the power spectral density of internal leakage acoustic signals and non-leakage noise signals, convolutional neural network(CNN) identification models of valve internal leakage were proposed. Aiming at the problem of large quantization error of physical theory and shallow network models in multi-conditions internal leakage data-sets, the deep belief network(DBN) regression model of valve internal leakage rate was proposed, and compared with traditional models such as support vector regression and back propagation neural network. The results show that the valve internal leakage diagnosis accuracy is 99% and the mean absolute percentage error(MAPE) of internal leakage rate quantification is 9.101 2, which proves the efficiency of the proposed models.