Aimed at the problem that traditional reliabilitybased optimization has poor efficiency for the huge ship finite element model, a new process is presented to improve efficiency and accuracy by synthesizing ensemble of surrogates with recursive arithmetic average (ER) and particle swarm optimization (PSO) into sequential optimization and reliability assessment (SORA). The uncertainty of variables is considered. The improved process is applied to the problem of reliabilitybased optimization of a multipurpose ship. SORA method decouples the process of reliability assessment and optimization and improves the efficiency of reliabilitybased optimization. PSO method guarantees the efficiency of calculating global optimal solution in the SORA. ER surrogate uses the minimal sampling points to establish a surrogate to meet with the accuracy requirement. Therefore, Kriging, RBF and SVR surrogates are combined to construct the ER surrogate which substitutes for timeconsuming process of calculating the most probable point (MPP) and optimization of SORA based on finite element model. The optimization results of the multipurpose ship show that reliabilitybased design optimization is more meaningful than deterministic optimization in practice. The improved process guarantees the high accuracy of solutions and significantly reduces the computational cost.