摘要
针对PVC管道漏水检测准确率不高和受外界影响的问题,提出一种可剔除异常数据,准确检测供水管道的漏水检测方法。首先利用自编码器的重构误差作为异常检测的阈值从而剔除异常数据,再经过长短期记忆网络与时间卷积网络改进的并行学习模型,来同步学习管道振动数据的长期依赖特征与局部特征,最后使用模型实现对漏水的检测。通过异常剔除与模型性能对比实验表明,该方法可以剔除异常值,使数据平稳化;相比于其他单一网络模型方法,检测准确率达到0.95,能准确识别漏水。
To solve the problems of low accuracy and susceptibility to the influence of the external environment in water leakage detection of PVC pipelines,this paper proposes a water leakage detection method for water supply pipelines that can eliminate outliers and achieve accurate detection.Firstly,the reconstruction error of the autoencoder(AE)is employed as the threshold for anomaly detection to eliminate outliers.Then,a parallel learning model improved by the long short-term memory network(LSTM)and temporal convolutional network(TCN)is adopted to synchronously learn the long-term dependency features and local features of pipeline vibration data.Finally,the model is adopted to realize water leakage detection.By conducting experiments on outlier elimination and model performance comparison,the results show that this method can eliminate outliers and make the data tend to stabilize.Additionally,compared with other singlenetwork model methods,the detection accuracy reaches 0.95,enabling accurate water leakage identification.
作者
许傲
郭改枝
尹志凌
张俊峰
XU Ao;GUO Gaizhi;YIN Zhiling;ZHANG Junfeng(College of Computer Science and Technology,Inner Mongolia Normal University,Hohhot 010022,China;Water Conservancy Career Development Center of Guyang,Baotou 014200,Inner Mongolia,China)
出处
《内蒙古师范大学学报(自然科学版)》
2025年第5期542-550,共9页
Journal of Inner Mongolia Normal University(Natural Science Edition)
基金
中央引导地方科技发展资金资助项目“‘人工智能+’管道漏水检测设备的实现”(2024ZY0144)
内蒙古自治区教育厅资助项目“多维度创新型研究生导师团队建设提升培养研究生创新能力的探索”(JG2024001Z)。
关键词
自编码器
异常值
长短期记忆网络
时间卷积网络
漏水检测
autoencoder
outlier
long short-term memory network
temporal convolutional network
water leakage detection
