To efficiently, accurately, and automatically achieve the optimized design of hollow structures, a chamber retaining wall was taken as the research object.Parametric modeling based on Python and ABAQUS was carried out, and the automation for the whole process of modeling, finite element calculation and analysis was realized. Combined with a chamber retaining wall in a sluice project, the multi-objective optimization design under multiple constraints was carried out by using non-dominated sorting genetic algorithm (NSGA-Ⅱ algorithm), aiming at the optimal of several strength and stability indexes and the minimization of concrete volume by taking the index threshold, geometric size and volume requirement as constraint conditions. In order to further improve the computational efficiency, a surrogate model was established based on artificial neural network algorithm, which could achieve a highly approximation to the actual finite element model. The results show that the proposed optimization process for chamber retaining walls can minimize concrete usage while effectively controlling strength and stability, and has significant economic benefits.