Objectives To break the monopoly of foreign patented technologies and ensure that the pump tower structure of LNG (liquefied natural gas) carrier cargo tanks meets the requirements for structural strength, fatigue life, and lightweight design under combined loading conditions, this study optimizes the key geometric parameters and sectional size variables of the pump tower.

Methods Surrogate modeling technology is integrated with heuristic optimization algorithms. First, a mathematical model for the pump tower optimization problem is established, followed by parametric modeling, multi-physical field loading, and initial sample selection using the optimal Latin hypercube sampling method. An XGBoost-based surrogate model is then developed, and an inexact adaptive sampling strategy is proposed to improve its accuracy. Subsequently, the surrogate model is coupled with four heuristic algorithms – NSGA-II, MOPSO, MOAHA, and NSWOA – to obtain the Pareto front. The TOPSIS method is used to compare the Pareto solution sets and determine a relatively optimal design, which is validated through direct finite element analyses.

Results The results show that the optimization scheme obtained by integrating the XGBoost model with the NSWOA algorithm is relatively optimal: the structural weight is reduced by 19.63%, and the fatigue life is improved by a factor of 1.503.

Conclusions A comprehensive and effective multi-objective optimization workflow is established for the mixed-variable design of the LNG carrier cargo tank pump tower structure based on surrogate models, which enhances design efficiency and provides a reference for the optimization design under combined loads.