基于梅尔频率倒谱系数特征识别供水管网漏损的机器学习模型比较研究  

A comparative study of machine learning models for identifying water supply network leakage based on Mel Frequency Cepstrum Coefficient features

作　　者：何立新 张宏洋 张峥 陈炯禧 王琦 管涌康[1,2] 龙岩 HE Lixin;ZHANG Hongyang;ZHANG Zheng;CHEN Jiongxi;WANG Qi;GUAN Yongkang;LONG Yan(School of Water Conservancy and Hydroelectric Power,Hebei University of Engineering,Handan 056038,China;Hebei Key Laboratory of Intelligent Water Conservancy,Handan 056038,China;School of Civil and Transportation Engineering,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]河北工程大学水利水电学院,邯郸056038 [2]河北省智慧水利重点实验室,邯郸056038 [3]广东工业大学土木与交通工程学院,广州510006

出　　处：《给水排水》2024年第8期137-143,共7页Water & Wastewater Engineering

基　　金：国家重点研发计划(2023YFC3209400);广州市重点研发计划农业和社会发展科技专题项目(2023B03J1333)。

摘　　要：当前,声信号在供水管网漏损识别领域备受关注,成为水务行业研究的焦点。针对探漏培训基地采集的声信号进行处理,提取梅尔频率倒谱系数特征,并运用支持向量机、随机森林、梯度提升决策树、XGBoost和BP神经网络五种有代表性的机器学习模型进行训练和测试。测试结果表明,5种机器学习模型都能有效识别管道中的漏损声信号特征,F1分数都超过86%。将上述模型应用于诊断实际管网中获取的漏损声信号,支持向量机表现最优,准确率达到82.8%,具有较强的泛化能力。结果验证了基于MFCC的机器学习模型可提高管网漏损诊断效率,降低维护成本。At present,the acoustic signal has attracted much attention in the field of water supply network leakage identification,and has become the focus of water industry research.In this paper,the acoustic signals collected from the leak detection training base are processed,and the characteristics of Mel Frequency Cepstrum Coefficient were extracted.Five representative machine learning models,including Support Vector Machine,Random Forest,Gradient Boosting Decision Tree,XGBoost and BP Neural Network,are used for training and testing.The test results show that the five machine learning models can effectively identify the characteristics of leakage acoustic signals in the pipeline,and the F1 Scores are more than 86%.The above model is applied to diagnose the leakage acoustic signal obtained in the actual pipe network.The Support Vector Machine performs best,with an accuracy rate of 82.8%,and has strong generalization ability.This paper verifies that the machine learning model based on MFCC can improve the efficiency of pipeline leakage diagnosis and reduce maintenance costs.

关 键 词：供水管网 漏损检测 声学信号 机器学习 MFCC 

分 类 号：TU990[建筑科学—市政工程]