DIRECT优化算法计算机仿真研究

DIRECT Optimization Algorithm by Computer Simulation

细分矩形方法(DIviding RECTangles,DIRECT)源于Lipschitz算法,其在优化过程中不断对中心点进行函数估值,因而迭代次数较多,并在不断采样过程中造成收敛速度缓慢等缺点.基于此,提出一种新的DIRECT算法,根据每次优化迭代产生的采样点来构建元模型,并识别最优域,在最优域集合上搜索最优点,因而加快算法的收敛速度.计算机仿真结果证明,RBF元模型,能明显加快DIRECT算法收敛速度,并能快速找到准确的全局最佳点,因而大大改善了DIRECT算法收敛性.

The DIRECT method was derived from Lipschitz algorithm,in the process of optimization,the central point was constantly valued by the method,so the number of iterations was more,and the convergence rate was slow in the process of continuous sampling.So a new DIRECT algorithm was put forward based on foregoing circumstances,the method used sampling points in each optimization iteration to construct metamodels and identify optimal region,in which the optimal point was searched,thus the method could speed up the convergence of algorithm.The computer simulation results showed that the RBF could significantly accelerate the convergence speed of DIRECT algorithm,and quickly find the global optimal point,thus greatly improving the convergence of the DIRECT algorithm.