基于CNN的水表故障检测算法研究

合格水表运行一段时间后可能出现硬件故障，造成水费计量异常。为避免此问题，传统上采用机器学习方法（例如支持向量机）分析水表日常读数以判断水表是否出现故障，但该方法常因人工选择特征不当而导致检测性能不能满足实用要求。为解决该问题，本文利用卷积神经网络（CNN）卓越的特征提取能力，根据水表日常读数自动提取水表故障特征，在此基础上提出一种基于CNN的水表故障检测方法，并通过大量实验对检测模型进行了参数优化。对比实验结果表明，本文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。

Research on CNN-based Fault Detection Algorithm for Water Meter

Qualified water meter may suffer from hardware fault after running for a period of time, which will lead to error in water selling charge. To avoid it, traditional machine learning method, such as support vector machine (SVM), is usually employed to analyze daily data of water meter and furthermore judge whether the water meter is running well or not. However, this method often fails to meet the practical requirements due to improper selection of artificial features. In order to solve this problem, by fulfilling the excellent ability of convolutional neural network (CNN) in feature extracting, this paper explores a CNN-based model to realize a novel fault detection method for water meter. Sufficient experiments also are carried out to optimize the CNN-based model. Finally, in comparison to SVM and integrated learning methods, our proposed algorithm shows apparently superior detection performance.