2种基于模式识别的环状燃气管网泄漏检测方法

Study on two kinds of leakage detection methods for loop gas pipeline network based on pattern recognition

为了将模式识别技术应用于环状燃气管网泄漏检测并找到合适的特征提取方法,以天津城建大学实验室环状燃气管网泄漏为例,将实验的28种工况作为测试样本,与之对应的模拟工况作为训练样本,采用提取压力图像特征向量法和节点压力矩阵法分别进行环状燃气管网的泄漏检测,采用支持向量机分类器将2种方法获得的特征向量进行训练与分类检验,进而将其分类准确率进行对比分析。研究结果表明:该2种方法均可用于环状燃气管网泄漏检测,提取压力图像特征向量法因有效地降低了特征向量的维度和数据波动的干扰,其结果更优。结合SCADA和GIS系统,可将该法应用于实际水、气、油管网泄漏检测和定位,有助于降低成本,提高检测效果。

In order to apply the pattern recognition technology to the leakage detection of loop gas pipeline network and find the suitable feature extraction method,the leakage of loop gas pipeline network in the laboratory of Tianjin Chengjian University was taken as an example. 28 kinds of experimental working conditions were taken as the testing samples,and the corresponding simulation working conditions were taken as the training samples. The ei-genvector extracting method from pressure images and the node pressure matrix method were applied respectively to carry out the leakage detection of loop gas pipeline network. The support vector machine（ SVM） classifier was used to conduct the training and classification testing on the eigenvectors obtained by the two methods,then the corresponding classification accuracy were compared and analyzed. The results showed that both the methods can be used for leakage detection of loop gas pipeline network. The eigenvector extracting method from pressure im-ages can effectively reduce the dimensionality of eigenvectors and the interference of data fluctuation,with the better results. Combined with SCADA and GIS system,the method can be applied to the leakage detection and lo-cation of the practical water,gas and oil pipeline networks,which can reduce the cost and improve the detection results.