Anomaly Detection Analysis Using AutoEncoder Based on Aids to Navigation Voltage Data

Aids to navigation (ATON) are essential facilities installed to support the safe navigation of vessels, providing critical navigational information. The electrical systems of ATON require a stable power supply, and any voltage anomalies must be promptly detected to ensure operational reliability. This study proposes an AI-based algorithm to address anomaly detection issues using voltage data collected from actual ATON systems between January 2018 and September 2021.BR/ In this study, unlabeled data without explicit fault labels were used, and anomalies were labeled using Isolation Forest, IQR-based labeling, and extreme value-based labeling methods. Based on these labeled datasets, the performance of anomaly detection was evaluated using AutoEncoder and Gaussian Mixture Model (GMM). Experimental results demonstrated that the AutoEncoder achieved an average F1 Score of over 83% with IQR and Isolation Forest-based labeling, showing superior performance. In contrast, GMM showed limited anomaly detection performance and struggled to effectively capture the complex patterns in the data.BR/ These findings highlight the potential for enhancing maritime safety and contributing to anomaly detection and fault prevention in various marine and industrial facilities.