Comparative analysis of multi-objective optimization algorithms with different surrogate models for trimaran side hull layout

This study formulates a multi-objective optimization method leveraging surrogate modeling to address the performance fluctuations of trimaran side-hull layouts under different working conditions, while circumventing excessive computational cost. Sample databases were constructed from CFD simulations for three representative speeds: low, medium, and high. Three surrogate models were developed: a Back Propagation Neural Network (BPNN), a Light Gradient Boosting Machine (LightGBM), and a Support Vector Machine (SVM). Grid search and K-fold cross-validation were used to perform hyperparameter optimization on these surrogate models. Building on these models, a single resistance minimization study was performed for three speeds by integrating surrogate models with Success-History Based Parameter Adaptation for Differential Evolution (SHADE), and the optimization performance of different surrogate models was systematically compared. Furthermore, using the weights of 0.2, 0.6, and 0.2 for low, medium, and high speeds, as an example, the Non-dominated Sorting Genetic Algorithm II (NSGA-II) was adopted to investigate the performance of three surrogate models in multi-objective optimization. The findings indicate that the optimization method based on the BPNN significantly reduces computational costs while maintaining prediction accuracy compared with the CFD validation results, and the proposed optimization framework demonstrates strong applicability and engineering potential for the optimized design of the trimaran.