基于改进自动编码器的水质数据异常检测研究  

Research on anomaly detection of water quality data based on improved autoencoder

作　　者：曹可欣 李永飞 韩博龙 王维政 CAO Kexin;LI Yongfei;HAN Bolong;WANG Weizheng(College of Computer,North China Institute of Science and Technology,Langfang 065201,Hebei,China)

机构地区：[1]华北科技学院计算机学院,河北廊坊065201

出　　处：《资源信息与工程》2025年第1期101-105,109,共6页Resource Information and Engineering

基　　金：河北省重点研发计划项目(19270318D);中央高校基本科研业务费资助项目(3142017067,3142023060);河北省物联网监控技术创新中心项目(21567693H)。

摘　　要：随着物联网监测设备在水资源领域的大量应用,监测数据呈爆发式增长,这些数据常常会面临各种干扰,产生大量异常数据,因此应利用异常检测算法对相关异常数据进行筛选。本文根据水质监测数据具有时空依赖性的特征,提出了一种基于卷积门控循环单元和自注意力机制的自动编码器模型,针对白洋淀某水域的水质监测数据进行异常检测,筛选出多项异常值。With the large number of applications of the internet of things monitoring devices in the field of water resources,the monitoring data has been growing explosively,and these data often face various interferences and produce a large amount of abnormal data,so anomaly detection algorithms should be utilized to screen the relevant abnormal data.In this paper,according to the characteristics of water quality monitoring data with spatial-temporal dependence,an autoencoder model based on convolutional gated recurrent unit and self-attention mechanism is proposed to detect anomalies for the water quality monitoring data of a water area in Baiyangdian,and screened out a number of anomalies values.

关 键 词：水质数据 异常检测 自动编码器 卷积门控循环单元 自注意力机制 无监督学习 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]