ICEEMDAN-adaptive multi-branch transformer for multi-engineering-scenario mooring tension prediction of floating platforms

Mooring line tension monitoring is critical for offshore floating platform safety, yet sensors in harsh marine environments suffer from damage and high maintenance costs. This paper proposes ICAMT, an ICEEMDAN-adaptive multi-branch transformer for economical and reliable mooring tension prediction. ICAMT comprises three components: an ICEEMDAN decomposition module for denoising and multi-scale feature extraction from unified multi-sea-state wave tank data; a MultiBranch-Transformer encoding full-, high-, and low-frequency IMF components to capture multi-timescale dynamics; and an IMFAA adaptive fusion module dynamically weighting branches for effective cross-branch information integration. To adapt to different scenarios, three prediction models are designed: DOF to single mooring tension prediction (ICAMT-D2ST), DOF to multiple mooring tension prediction (ICAMT-D2MT), and single mooring tension to multiple mooring tension prediction (ICAMT-T2MT). Key configurations are determined through parameter optimization, and comparisons with various deep learning baselines validate the effectiveness and superiority of the proposed method. Ablation experiments confirm module contributions; comparisons with decomposition methods such as CEEMDAN, EEMD, EMD, and VMD demonstrate the superiority of ICEEMDAN in mooring tension prediction tasks and the effectiveness of modal decomposition methods. The method is trained and evaluated on a unified dataset covering multiple sea-state conditions, providing a feasible technical solution for mooring tension monitoring in the evaluated conditions.