A comparative study of machine learning models for identifying water supply network leakage based on Mel Frequency Cepstrum Coefficient features
At present,the acoustic signal has attracted much attention in the field of water sup-ply network leakage identification,and has become the focus of water industry research.In this pa-per,the acoustic signals collected from the leak detection training base are processed,and the char-acteristics 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 diag-nose 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 leak-age diagnosis and reduce maintenance costs.
Water supply networkLeakage detectionAcoustic signalMachine learningMFCC
万方数据
维普
