Abstract
With the rapid development of maritime transportation, the issue of safe navigation for ships in complex and ever-changing marine environments has attracted increasing attention. Hull girder load is regarded as an important indicator of wave-induced loads during ship navigation and its prediction plays a crucial role in ensuring hull structural safety. To overcome the time-consuming nature of nonlinear load predictions in viscous flows caused by fluid-structure interaction, a data-driven load prediction model is proposed by combining Bayesian optimisation (Bo), a Long Short-Term Memory neural network (LSTM) and a Transformer algorithm. In our research, the navigational status of a S175 ship is simulated using the CFD-FMA method. These navigational data are then used to establish the Bo-Transformer-LSTM model to predict the load fluctuations along the ship’s length. The impacts of factors such as input sources, prediction ranges, and wave cases are investigated for this hybrid model. Its advantage is also illustrated by comparing it with the traditional data-driven algorithms, such as BP, RBF and LSTM. In addition, the wider practicality of the hybrid model is further verified by the application of the transfer learning strategy. This research can provide a reliable rapid prediction approach for wave load assessment and contribute to the optimal design of hull structures.