基于平滑阈值与孤立森林的大坝监测数据异常检测  

Anomaly Detection of Dam Monitoring Data Based on Smoothing Threshold and Isolated Forest

作　　者：张瑜 秦学 彭浩 ZHANG Yu;QIN Xue;PENG Hao(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China;Guiyang Survey Design and Research Institute Co.,Ltd.,Power Construction Corporation of China,Guiyang 550081,China)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025 [2]中国电建集团贵阳勘测设计研究院有限公司,贵州贵阳550081

出　　处：《人民黄河》2025年第3期141-145,共5页Yellow River

基　　金：贵州省科技计划项目(黔科合支撑[2023]一般251)。

摘　　要：为解决孤立森林算法检测大坝异常数据时因不能识别数据间趋势性和相关性而造成数据误判的问题,提出基于平滑阈值与孤立森林的大坝监测数据异常检测算法。首先利用小波变换提取时序数据的趋势项,然后使用ARMA模型对提取的趋势项数据确定动态阈值区间,最后利用孤立森林算法检测出散落在阈值区间外的异常值。以贵州省毕节市夹岩水利枢纽工程混凝土面板堆石坝为例,分别对大坝坝基、坝体、周边缝及面板4个部位监测数据进行检测,验证算法效果。结果表明:与传统孤立森林算法相比,基于平滑阈值与孤立森林的算法对压力、观测房沉降量、开合度、应力的误判率分别降低了12.2、13.4、7.1、8.0个百分点。In order to solve the issue of data misjudgment caused by the inability to identify the trend and correlation between data when the isolated forest algorithm detected dam abnormal data,an anomaly detection algorithm based on smooth threshold and isolated forest was proposed.Firstly,the trend term of time series data was extracted by wavelet transform.Secondly,ARMA model was used to determine the dynamic threshold interval of the extracted trend item data.Finally,isolated forest algorithm was used to detect the outliers scattered outside the threshold interval.Taking the concrete faced rockfill dam of Jiayan Key Water Control Project in Bijie City,Guizhou Province as an example,the monitoring data of four parts of the dam foundation,dam body,peripheral joint and panel were tested respectively to verify the effective⁃ness of the algorithm.The results show that comparing with the traditional isolated forest algorithm,the algorithm based on smooth threshold and isolated forest reduces the misjudgment rates of pressure,observation room settlement,opening and closing degree,and stress have been reduced by 12.2,13.4,7.1,and 8.0 percentage points,respectively.

关 键 词：小波变换 ARMA模型 孤立森林 异常检测 大坝 毕节市夹岩水利枢纽工程 

分 类 号：TV698.1[水利工程—水利水电工程]