Outlier Detection of Gravity Dam Deformation Monitoring Data Based on the Multiple Local Outlier Coefficient Method

Abstract

In the deformation monitoring data of gravity dams, a few outliers are often found, which will have adverse effects on the monitoring of model building and other data analysis work. In this study, the multiple local outlier coefficient method was proposed to quickly detect the outlier data in real time and to provide high-quality data for subsequent data analysis. This method was based on the idea of distance in the outlier detection algorithm, aiming at laws and characteristics of gravity dam deformation monitoring data. First, the basic principle, calculation steps, and basic features of the multiple local outlier coefficient method were studied. Then, for the two important parameters of the algorithm, the appropriate window length was selected using autocorrelation and partial autocorrelation analysis, and the appropriate threshold values were selected using the 3σ criterion, maximum method, and empirical method. Finally, an engineering example was used to verify that the algorithm could accurately detect the outliers in the gravity dam deformation monitoring data, and the deviation degree and meaning of the outliers were understood according to the calculated outlier coefficient. The multiple local outlier coefficient method has the advantages of a simple calculation principle, fast calculation speed, real-time detection, and a clear meaning of calculation results. By selecting the appropriate parameters, the method could satisfy the outlier detection of different types of data, offering an advantage in adapting to the computing demand of massive monitoring data and improving the intelligence and real-time monitoring of dam safety.