基于多任务卷积神经网络的供水管道泄漏识别和定位方法  

Leak detection and localization using multi-target convolutional neural networks

作　　者：张鹏 赫俊国 黄婉仪 张杰 袁永钦 袁一星 陈波 ZHANG Peng;HE Junguo;HUANG Wanyi;ZHANG Jie;YUAN Yongqin;YUAN Yixing;CHEN Bo(School of Environment,Harbin Institute of Technology,Harbin 150090,China;School of Civil Engineering,Guangzhou University,Guangzhou 510006,China;Guangzhou Water Supply Company Limited,Guangzhou 510160,China;Hunan Puqi Water Environment Institute Co.,Ltd.,Changsha 410201,China)

机构地区：[1]哈尔滨工业大学环境学院,哈尔滨150090 [2]广州大学,广州510006 [3]广州市自来水有限公司,广州510160 [4]湖南普奇水环境研究院有限公司,长沙410201

出　　处：《给水排水》2023年第8期135-144,共10页Water & Wastewater Engineering

基　　金：广州市科技计划(202103000098)。

摘　　要：针对传统地面声学技术过度依赖工人经验的弊端,基于多通道信号的多任务卷积神经网络(MTCNN),开发了应用该模型的无线多探头漏损定位仪。多任务卷积神经网络模型模型结合无线多探头漏损定位仪,能够地面上探测泄漏并且定位漏点位置。在实际管道的应用结果表明,所提出的方法的识别准确率达98.63%,定位的平均误差为0.2 m,效果显著。Ground acoustics technology is a widely used and effective method for detecting water leakage in supply pipe networks.To address the limitation of traditional ground acoustics technology,which overly relies on the expertise of workers,this study developed a wireless multi-probe leak locator based on the multi-task convolutional neural network(MTCNN) with multi-channel signals.The MTCNN model,combined with the wireless multi-probe leak locator,can detect leaks on the ground and locate the leak points.The application results in actual pipelines show that the proposed method achieves a recognition accuracy of 98.63%,with an average location error of 0.2 meters,demonstrating significant effectiveness.

关 键 词：供水管道 泄漏检测 机器学习 漏点定位 多任务学习 

分 类 号：TU990[建筑科学—市政工程]