Abstract: Kriging surrogate model can effectively approximate the engineering optimization problem with black box constraints.However,there exists poor modeling accuracy and low efficiency of engineering optimization based on initial sample data.Considering above problems,a new effective global optimization algorithm was proposed,which combined the Kriging model,the weighted expectation improvement criterion and the evolutionary algorithm.Inspired by the expectation improvement criterion,this new filling criterion constructed the weight function to adjust the global and local exploration ability according to the weight function and the expected increment,so that it had the characteristics of self-adaptive adjustment according to the distance depended new sample point and then turned the local optimal solution to global optimization.Numerical and engineering cases showed that the global optimization algorithm under the new criterion and Kriging model could achieve fast solution to the optimization problem with high accuracy and good stability.

Key words: Kriging model, weighted expectation improvement criterion, global optimization problem, probability of feasibility

摘要： Kriging代理模型可有效近似包含黑箱约束的工程优化问题,但存在仅依靠原始样本数据建模精度不高,工程优化效率低等问题。针对上述问题,提出了综合Kriging模型、权期望改进准则及进化算法的新有效全局优化算法。该算法的权期望填充准则在期望改进准则启发下,依据距离函数与期望增量的函数关系,构造权函数实现对新填充准则全局及局部探索能力的调整,使其具有依据试验点距离进行自适应调整进而跳出局部最优解实现全局优化的特性。数值算例和工程实例结果表明,在新准则和Kriging模型作用下的全局优化算法能够实现对优化问题的快速求解,精度高且具有较好的稳定性。

关键词: Kriging模型, 权期望填充准则, 全局优化, 可行性概率