Intelligent prediction of wave loads based on multi-source data-driven methods

High-fidelity wave-load tests is pivotal in the design and development of modern ship structures. Nonetheless, the prohibitive cost of conducting such tests limits their applicability. A wave-load data fusion method was implemented using the model test data from a typical displacement-type ship to generalise the small sample test, and a multi-source data characterisation and model parameter optimisation method were established to achieve efficient wave-load (wave frequency and slamming superposition) prediction. Using multi-source data, a multi-fidelity wave-load prediction (MFWLP) model was constructed by combining numerical computation, regular wave tests (RWT), and irregular wave test (IWT) data based on transfer learning with a secondary correction wave load prediction under irregular wave conditions. The model was effectively employed in a ship wave-load engineering case. This study reveals that the adopted MFWLP model demonstrates adequate extrapolation and prediction capabilities, even with limited high-fidelity test samples. The discrepancy between the nonlinear wave load prediction and model test was found to be within 15%. Therefore, the proposed method can serve as a foundation for rapidly assessing ship wave loads across various sea conditions, especially under high sea states.