Experimental research On CEEMDAN-GA-VMD-Transformer ultra-short-term prediction of offshore wind turbine operation and maintenance ship attitude base on wind-ship coupling motion feature-driven

Accurate and real-time ship attitude prediction is crucial for developing wave compensation systems and ensuring the safety of offshore wind turbine operation and maintenance. Under the coupling effect of wind, wave and hull, ship attitude exhibits complex spatio-temporal dynamic characteristics, which are difficult for traditional methods to predict. In this paper, CEEMDAN and GA-VMD are adopted to realize adaptive dual modal decomposition, which effectively reduces modal complexity and aliasing, and improves model feature extraction ability. Wind speed and direction characteristics are selectively fused with IMFs using the Pearson correlation coefficient, and Transformer is used for modeling. Experiments show that the prediction errors of roll, pitch and heave of the model proposed in this paper remain below 5.57% across different prediction steps and different sea conditions, providing reliable online support for the development of wave compensation systems. • Proposes an adaptive dual decomposition strategy to reduce modal aliasing, enhancing ship attitude feature extraction. • Develops a wind-ship coupling fusion mechanism, cutting prediction errors by 28.5% under high sea conditions via dynamic weighting. • Transformer captures multi-scale temporal dependencies, achieving less than 5.57% errors in all predictions across steps. • Multi-scenario validation shows minimal error growth under varying steps and high sea states, supporting active wave compensation systems.