케이슨식 방파제 대상 실시간 모니터링 데이터 분석을 위한 온라인 학습 기반 적응적 이상상태 탐지 알고리즘 개발

Development of Online-learning based Adaptive Anomaly Detection Algorithm for Monitoring Data Analysis on Caisson Type Breakwater

Most port structures are massive and data measured on them sensitively changes to the surrounding environment including sea waves, tides, wind, and other operational conditions so it might be difficult to extract and long-term monitor their own features such as natural frequencies and mode shapes. To solve this problem, an anomaly detection algorithm with online learning was developed for the analysis of monitoring data on the port structures. For this, data were first measured on a 1/50 scaled model of caisson type breakwater through hydraulic model experiments, and the characteristics of data were investigated. Then an unsupervised algorithm was developed to online detect abnormal conditions caused by the drift, which can track the reconstruction error from the principal component analysis and the Euclidean distance between original and reconstructed signals. The experimental results showed that the proposed algorithm could be successfully applied to time-dependent dataset shifts with high accuracy and automatically calculate the threshold based on the adaptive model.