Ship unsinkability is an important performance to evaluate the ship survivability, and it is also a key index for optimizing the subdivision strategy. However, the high cost is still a difficulty restricting the practical application of the unsinkability optimization. More effective methods will be provided with the continuous application of machine learning technology. The optimization of unsinkability subdivision is solved by using the particle swarm optimization (PSO) algorithm based on reinforcement learning. And the optimization system development and interface design is implemented for the integrated machine learning module and unsinkability design module. The effect of different parameter settings in the algorithm on the optimization efficiency is also discussed. The analysis of the optimal solution shows that this method can effectively find out the optimal subdivision scheme. It can provide a basis for formulating scientific subdivision strategies.