基于卷积神经网络的水表故障检测算法  

CNN-based fault detection algorithm for water meter

作　　者：徐艺文[1] 王芝燕 李立春 李贵生 XU Yiwen;WANG Zhiyan;LI Lichun;LI Guisheng(College of Physics and Infoimation Engineering,Fuzhou University,Fuzhou,Fujian 350108,China;Zhiheng Technology Company Limited,Fuzhou,Fujian 350109,China)

机构地区：[1]福州大学物理与信息工程学院,福建福州350108 [2]智恒科技股份有限公司,福建福州350109

出　　处：《福州大学学报（自然科学版）》2020年第3期314-317,324,共5页Journal of Fuzhou University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61671152);住房和城乡建设部科技资助项目(2017-R2-016)。

摘　　要：利用卷积神经网络(CNN)提取水表故障特征,提出一种基于CNN的水表故障检测方法,并通过大量实验对检测模型进行参数优化.对比实验结果表明,所提方法相比于支持向量机和集成学习方法,具备更高的检测性能,且检测精度满足实用需求.By convolutional neural network(CNN),fault features of water meter were extracted.Then a CNN-based model was explored to realize a novel fault detection method for water meter.Sufficient experiments also were carried out to optimize the CNN-based model.Finally,in comparison to support vector machine(SVM)and integrated learning methods,our proposed algorithm showed apparently superior detection performance.

关 键 词：故障检测 水表 卷积神经网络 特征提取 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]