高级检索+

基于数据重构与孤立森林法的大坝自动化监测数据异常检测方法

An Anomaly Detection Method for Dam Automatic Monitoring Data Based on Data Reconstruction and Isolated Forest

  • 摘要: 大坝安全自动化采集的监测数据不可避免地存在粗差、缺测等问题,针对异常数据量值、数量以及分布规律的不确定性,人工删除异常值存在工作量大、主观性强等不足,提出一种基于数据重构与孤立森林法的异常数据检测方法,该方法是一种无监督的学习方法,不需要根据特征标签进行样本学习,适用范围较广。首先对大坝自动化监测数据进行分解与重构,分离出趋势项,而后用孤立森林算法对剩余项进行判别,计算测点的异常分数,并剔除明显的异常数据,最后再根据拉依达准则进一步清理异常数据。通过实例验证,该方法能较好检测出大坝安全自动化异常监测数据,满足工程实际应用。

     

    Abstract: Because the outliers are inevitable in dam automation monitoring data,and the magnitude,quantity,and distribution of outliers are uncertain. So the manual deletion of outliers has a large workload and strong subjectivity. This method has a wide range of applications because it does not require sample learning based on feature tags. First of all,the trend item is extracted after the automatic monitoring data is decomposed and reconstructed. Then the obvious outliers in the remaining item are discriminated and eliminated by using the isolated forest algorithm. Finally,the other outliers are performed according to the Pauta criterion. Through the example verification,the method can meet the practical application of the project which can effectively detect the outliers of the dam automatic monitoring data and the false positive rate is low.

     

/

返回文章
返回