基于液位监测及CNN-SVM的排水管网缺陷诊断

Fault Diagnosis Method of Drainage Network Based on Liquid Level Monitoring Data and CNN-SVM

为提高排水管网运维管理能力,使管网结构性、功能性缺陷问题得到有效监测并识别,通过分析排水管网监测任务的具体要求,提出了一种基于液位监测数据及CNN-SVM的排水管网缺陷诊断方法,通过将SVM分类器替换Softmax分类器以改善CNN的分类性能,同时规避SVM对于数据特征提取的劣势。针对排水管道监测环境的复杂性,自行设计并搭建排水管道缺陷试验装置,并结合物联网监测系统进行数据的采集。结果显示,模型能十分有效地进行排水管道缺陷问题的诊断排查,在十分类、十三分类、全分类任务下分别具有94.20%、91.57%、85.34%的准确率。与其他诊断模型相比,在分类精度要求最高的全分类任务中CNN-SVM模型的准确率比次优的CNN-LSTM模型高出了16.94%,并且在精确率、召回率、F1-Measure上也具有明显优势,验证了所提模型的泛化性和有效性。

In order to improve the operation and maintenance management ability of drainage pipe network,and the structural and functional defects of pipe network can be effectively monitored and identified,a fault diagnosis method of drainage network based on liquid level monitoring data and CNN-SVM was proposed by analyzing the specific requirements of drainage network monitoring tasks.Softmax classifier was replaced by SVM classifier to improve the classification performance of CNN and avoid the disadvantages of SVM in data feature extraction.In view of the complexity of drainage pipeline monitoring environment,the drainage pipeline defect test device was designed and combined with the Internet of Things monitoring system to collect data.The results showed that the model was very effective in the diagnosis and troubleshooting of drainage pipe defects,with an accuracy of 94.20%,91.57%and 85.34%under the tasks of ten classification,thirteen classification and full classification,respectively.Compared with other diagnostic models,the CNN-SVM model had a 16.94%higher accuracy than the second‑best CNN-LSTM model in all classification tasks requiring the highest classification accuracy,and also had obvious advantages in accuracy rate,recall rate and F1-Measure,which verified the generalization and effectiveness of the proposed model.