Hyungtak Ju, Kyoungwon Park, Juhyoung Sung, Kiwon Kwon, Byoungchul Song, Taeho Im, Journal of Internet Computing and Services, Vol. 26, No. 5, pp. 65-74, Oct. 2025
10.7472/jksii.2025.26.5.65, Full Text:  HTML
Keywords: Ship Stability, Ship Safety, Time-Series Forecasting, Transformer-based Models

Abstract

Dynamic assessment of ship stability is essential for ensuring navigational safety and preventing maritime accidents in unpredictable sea conditions. Conventional static stability analysis cannot capture complex interactions among multi-degree-of-freedom motions such as roll, pitch, and heave in real ocean environments. This study proposes a hybrid prediction framework integrating high-fidelity dynamic GZ (righting arm) time-series data from physics-based simulators with advanced deep learning models. Transformer-based architectures specialized for long-term forecasting—Informer, Autoformer, and TimesNet—were applied to predict future GZ values, with performance evaluated using RMSE and MAE metrics. TimesNet achieved the best performance with the lowest RMSE (0.0358 m). In risk detection evaluation linking predicted GZ values with safety thresholds (SAFE/UNSAFE), TimesNet uniquely achieved precision of 1.0 at the 9° threshold, demonstrating high reliability in detecting hazardous situations. This research establishes a foundation for real-time ship stability assessment by combining physics-based simulation and data-driven prediction, presenting potential as a core safety module for autonomous ships.

Statistics
Cite this article