Data-Driven Study on Dynamic Tension Response and Intelligent Prediction in Deepwater Mooring Systems

Le résumé fourni par la source

Abstract Deepwater semi-submersible platforms operating in complex sea states exhibit mooring-line tensions with pronounced nonlinearity and multi-scale fluctuations. In engineering practice, reliable and fast tension prediction is critical for online integrity assessment and operational decision support, particularly when sea states vary and load redistribution may occur after a single-line failure. Although high-fidelity fully coupled environment-platform-mooring simulations can deliver reliable responses, their computational cost is high, which limits their use for on-demand online assessment. This paper targets representative sea states in the South China Sea and develops a time-domain coupled model for the environment, platform (floater), and mooring system. OrcaFlex simulations are driven by measured sea-state parameters and the platform six-degree-of-freedom (6-DOF) motion responses to obtain effective tension time histories for 16 mooring lines. Based on these results, sensitivity analyses on polyester segment length and diameter are conducted, and load redistribution after a single mooring-line failure is investigated to identify key governing parameters and higher-risk headings for the taut mooring system. A two-layer long short-term memory (LSTM) model is then developed for one-step-ahead prediction using a 100 s sliding window at 1 Hz. To avoid time-series information leakage, complete one-hour windows are split into development and independent test subsets, yielding 17,505 samples in total. The results show that a 10% reduction in line length increases peak tension by approximately 65%, while a 20% reduction in diameter increases peak tension by about 24%. After a line failure, maximum tensions of adjacent-heading lines increase by about 25%, whereas those in symmetric headings decrease by about 12%. For all 16 lines, the LSTM achieves RMSE of 63.9-150.4 kN with R2 of 0.964-0.993 and Acc of 0.95-0.978, and it consistently tracks both low-frequency drift and higher-frequency fluctuations with limited phase error. Compared with a simple MLP baseline, the LSTM markedly reduces peak-region deviations and lagging behavior. The proposed workflow provides a reusable framework for deepwater mooring parameter design, post-failure condition assessment, and rapid tension prediction.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.