Objectives  In view of the frequent occurrence of marine accidents caused by low ultimate bearing capacity of ships. To this end, the improved AK-MCS method is proposed for the research of the ultimate strength reliability of ship structures.

  Methods  The information entropy function H was introduced to perform secondary optimization on the sample points to improve the quality of the best sample points, so as to improve the accuracy and update efficiency of the Kriging model; and then the iterative stop criterion of AK-MCS was replaced by K-fold cross validation, so as to avoid over-learning and under-learning situation, thus the Kriging model is trained with fewer sample points to achieve a high degree of fitness to the limit state function. The nonlinear oscillator mathematical model is used to verify the results. The improved AK-MCS method is applied to the ultimate strength reliability study of shipboard frame.

  Results  The results show that the number of calls to the finite element calculation in the improved AK-MCS method is reduced by 38% compared with the original method, and the efficiency and accuracy of the improved AK-MCS method are verified. The application results show that The calculation accuracy of this method is the highest, and the number of calls to finite element calculation is reduced by 32% compared with the original method. This verifies the applicability and efficiency of the improved AK-MCS method in the study of the ultimate strength reliability of ship plates.

  Conclusions  The improved AK-MCS method is used to research the ultimate strength reliability of ship plates and evaluate the failure probability of local structures, especially dangerous structures during actual navigation.