Outlier Detection Method of Dam Monitoring Data Based on Robust Estimation and Variable Separation

Abstract

The original monitoring data of dams is the most important data to grasp the operation behavior of the dams,and the outliers in the data are the focus during the analysis.Outliers are divided into two categories.One category is caused by measurement errors and should be eliminated or supplemented to avoid affecting subsequent analysis.The other is caused by structural mutations and should be highly valued.At present,main outlier recognition methods in dam engineering are based on traditional mathematical statistics and do not consider the influence of structural anomalies,which results in low recognition accuracy.Therefore,based on an in-depth study of dam monitoring data and outlier characteristics,this paper first employs robust MM estimation to eliminate the normal influence of internal and external factors and then adopts the residual measured value to eliminate the stable abnormal influence by difference before and after.Finally,according to the minimum value method,outlier identification is conducted on the residual values.The application of the measured dam data proves that the proposed method can identify the measurement outliers more effectively and robustly,and avoid the interference of structural stability anomalies.

Keywords

• outlier detection;time series data;robust estimation;dam monitoring;variable separation