提出一种应用于黑盒问题（ black⁃box problem ）的优化算法，称为遗传算子采样（ genetic operator sampling，GOS）自适应代理优化算法。通过对当前样本进行两两交叉，对当前最优样本进行高斯变异2种算子获得候选样本集。对候选样本进行适应性评估，评估标准为候选样本处的交叉验证误差和该样本与父代样本之间最小距离的乘积，将乘积最大的样本加入已有样本集。 GOS优化算法在一维问题中详细阐述，与有效全局优化算法（ efficient global optimization，EGO）和最大化模型误差算法（ maximum square error，MSE）在3个典型数学算例中进行对比，验证该算法的有效性。%This paper proposes an optimization algorithm that is applied to Black⁃box Problem, called Genetic Op⁃erator Sampling ( GOS ) adaptive surrogate⁃based optimization algorithm. Genetic operators produce candidate sample set. Cross⁃over operator is executed between any two of samples and mutation operator is executed only on present best sample. Then an assessment criterion, which is the product of the cross validation error of the candidate sample and the minimum distance between it and existing samples, is used to judge the adaptation of each sample. The candidate sample with largest product will be added to existing samples. GOS is illustrated on 1⁃D function in detail and is compared to EGO and MSE algorithm on three typical functions, the results validated the effectiveness of GOS algorithm.
展开▼
