댐 계측데이터 분석의 자동화를 위한 계측기기별 이상치 탐지 방법 결정을 위한 연구

초록

Dams, special bridges, tunnels, and other social infrastructure play a crucial role in national water resource management, maintaining traffic flow, ensuring safe transportation, and forming the backbone of the national economy and public life. These infrastructures require periodic precise safety inspections to continuously assess their structural stability and functional reliability. However, the analysis of measurement data required for preparing detailed safety inspection reports involves a vast amount of data and high complexity, necessitating significant manpower and time. In particular, the preprocessing stage of detecting and removing outliers in measurement data is prone to errors, reducing work efficiency. To address these challenges, the automation of measurement data analysis is necessary. An automated measurement data analysis system enhances the consistency and efficiency of report generation while reducing labor and costs. Therefore, this study focuses on techniques for detecting and removing outliers in raw data as a step toward establishing an automated measurement data analysis system. Univariate and multivariate outlier detection methods were reviewed and compared. Based on the characteristics of each instrument, the most appropriate detection technique was selected. For pore water pressure gauges, the Local Outlier Factor (LOF) method was identified as the most suitable approach, showing the greatest improvement in correlation after outlier removal. The proposed methodology can be applied to other instruments as well and will serve as a foundation for improving the reliability and precision of automated monitoring data analysis systems.

키워드