Surrogate Model-based Multi-Objective Optimization in Early Stages of Ship Design

The abstract is the early stages of ship design, decision ship's main dimensions significantly impacts performance and total cost ownership. This paper focuses on an optimization approach based surrogate models at design. objectives are to minimize power requirements building costs while still satisfying constraints. We compare three approaches models: Kriging, BPNN-PSO (Backpropagation Neural Network-Particle Swarm Optimizer), MLP (Multi-Layer Perceptron) in two multi-objective algorithms: MOEA/D (Multi-Objective Evolutionary Algorithm Decomposition) NSGA-II (Non-Dominated Sorting Genetic II). experimental results show that get best with MAE 6.03, gets second position 7.2. improve design around 58% smaller adequate 6% less steel weight than original However, have lower hypervolume Kriging for both algorithms NSGA-II. On other hand, has most inadequate model accuracy performance, 22.2, but produces highest hypervolume, lowest computational time, far objective values algorithms, Nevertheless, can solutions reduce work time