一种新颖的混合响应面优化方法

Novel hybrid response surface optimization method

为减少黑箱优化过程中的评估次数,提出了一种新颖的混合响应面优化方法(HRSO),利用混合响应面建立高精度的近似模型作为代理模型,通过迭代更新响应面不断接近真实模型,从而完成优化。以Dixon-Szego函数类作为测试函数,以评估次数为方法性能优劣的评价指标,实验结果表明,与Gutmann-RBF、CORS-RBF两种方法相比,HRSO能够在较少的评估次数内满足相同的收敛条件,且向全局快速收敛,是一种适合求解黑箱优化问题的方法。

This paper presented a novel method,called hybrid response surface optimization（HRSO） to reduce the number of evaluations in the process of optimization of black-box functions.The proposed method used hybrid response surface to build high accuracy black-box approximate models as surrogate-models.Then it updated the approximate model by circles of iteration.This paper applied the method on the Dixon-Szego test functions and estimated the performance by the number of function evaluations,when a run satisfied the convergence criteria.The results indicate that HRSO meets the same convergence by less evaluation comparing to Gutmann-RBF and CORS-RBF.And it can converge to global optimum quickly.It is a suitable method for solving expensive black-box problem.