Abstract: Aiming at the failure probability analysis of cylindrical structures under underwater explosion load, a surrogate model was established by combining BPNN optimized by genetic algorithm and Monte Carlo Simulation, and hybrid learning function was adopted to find the new training point for model updating. The calculation efficiency and accuracy of the algorithm were verified by numerical examples, and the high-efficiency and high-precision analysis of cylindrical shell structures under underwater explosion load was realized. The results show that the BPNN optimized based on genetic algorithm combined with Monte Carlo Simulation method can significantly improve the computational efficiency, and can be applied to the reliability analysis of cylindrical shell structures under underwater explosion load. The research results have reference significance for the risk assessment and safety-based design of structures.