In order to design a marine internal pressure resistant square cabin which meets the requirements for strength and lightweight design, the neural network surrogate model is combined with heuristic intelligent optimization algorithms and applied to the shape and size optimization of the components of such a cabin.

The corner chamfer radius, plate thickness and beam model number are selected as design variables for conducting three-dimensional parametric modeling, and sample points are selected according to the optimal Latin hypercube experimental design method. The response values of these sample points are then calculated to build a radial basis functions (RBF) neural network surrogate model. To perform global optimization, the surrogate model is combined with three heuristic optimization algorithms respectively: an adaptive simulated annealing algorithm (ASA), multi-island genetic algorithm (MIGA) and particle swarm optimization (PSO) algorithm.

The results show that the three hybrid optimization methods can all reduce structural weight on the basis of meeting the allowable strength requirements, and the optimal solution sought by the RBF-ASA method in the overall situation has a relatively good weight reduction effect.

This study can provide valuable references for the optimal design of internal pressure-resistant square cabin structures, giving it great significance for overcoming the key technical problems faced by ships using nuclear power plants.