Abstract: Aiming at the optimization design problem of complex large-scale engineering structure under parallel simulation technology,combining adaptive surrogate model optimization and computer parallel computing technology,a surrogate model optimization algorithm based on constrained parallel adaptive sampling was proposed.The sampling method of the algorithm was composed of a local minimum model prediction single-point strategy and a global dual-objective constraint multi-point strategy.The sample points were selected by constructing the constraint expectation improvement function and the sample space sparsity function,so that the obtained new sample points had the ability to balance the search for the local optimal region of the objective function and the development of the global feasible boundary.The comparison and analysis of test examples and existing algorithms showed that the algorithm had better optimization efficiency,optimization accuracy and stability.Finally,the algorithm was applied to the multi-parameter optimization of the large steel structure radial gate,and three kinds of adaptive surrogate model optimization algorithms and genetic algorithm based on static surrogate model were used to solve the problem respectively.The results showed that the proposed algorithm had achieved a more significant optimization effect under the conditions of working performance and safety,the gate quality reduced by 42.85%,which maximized the material performance and saved the gate cost.

Key words: adaptive surrogate model, parallel optimization algorithm, multi-point strategy, constraint optimization, radial gate optimization

摘要： 针对并行仿真技术下复杂大型工程结构的优化设计问题,将自适应代理模型优化和计算机并行计算技术相结合,提出一种基于约束并行自适应采样的代理模型优化算法。算法的采样方法由局部最小模型预测单加点策略和全局双目标约束多加点策略构成,通过构造约束期望提高函数和样本空间稀疏度函数对样本点进行筛选,使所获得的新样本点兼顾搜索目标函数局部最优区域和开发全局可行边界。通过对比分析测试算例与已有算法表明,该算法具有更好的优化效率、优化精度和稳定性。最后将算法运用于大型钢结构弧形闸门结构的多参数优化,分别采用3种自适应代理模型优化算法与基于静态代理模型的遗传算法进行求解,验证了所提算法的优越性。

关键词: 自适应代理模型, 并行优化算法, 多加点策略, 约束优化, 弧形闸门优化