Due to the complexity of ship structure and collision optimization, traditional optimization methods are difficult to effectively use.Here, a new structural collision resistance performance optimization method, i.e., SGA-BP-GA, was proposed based on genetic algorithm (GA), simulated annealing algorithm and BP neural network combined with the orthogonal test design and ABAQUS parametric simulation technique.In order to improve prediction accuracy and generalization ability of the BP network for the structural crashworthiness index, the probabilistic jump feature of the simulated annealing algorithm was used to overcome the genetic algorithm’s shortcomings of being easy to be precocious and trapped in local optimization.Then, the simulated annealing genetic algorithm (SGA) was used to optimize weights of the BP network.The proposed SGA-BP-GA method was used to optimize the collision resistance performance of a FPSO’s side structure to verify its correctness and feasibility.The results showed that compared to traditional BP, GA-BP and SA-BP, the SGA-BP one has higher prediction accuracy and generalization ability; compared to the GA-BP-GA method, the optimized results of the SGA-BP-GA one increase 5.34%; the proposed SGA-BP-GA method is more appropriate for the complex optimal design of ship structures’ collision resistance performance.