摘要
在谱尺度BFGS算法基础上提出了一种扰动谱尺度BFGS算法,即在谱尺度BFGS算法的矩阵迭代公式中加入一个扰动因子,该因子能保证该算法求解非凸函数极小值问题时具有全局收敛性.在求解大规范问题时,该算法也能改善拟牛顿矩阵条件数,从而降低求解子问题的难度.通过数值试验对该算法进行检验,结果表明:在相同条件下,求解大规模问题时,该算法优于谱尺度BFGS算法.
A Perturbed SSFGS method was proposed on the basis of SSBFGS method, namely, a perturbed factor introduced into update formula of SSBFGS method. The perturbed factor guarantees that this method is grobally convergent for nonconvex minmiza- tion problem. When it is used to solve the large scale problems, the method can improve the condition number of quasi - Newton matrix so that the related subproblem is relatively easy to be solved. Finally, some numerical experiments are done to test the pro- posed methods. The results show that when it is used to solve large scale problems, the Perturbed SSBFGS method outperforms the SSBFGS.
出处
《宜宾学院学报》
2013年第12期34-37,共4页
Journal of Yibin University
关键词
非凸函数极小值
谱尺度BFGS算法
全局收敛性
nonconvex function minimization
spectral- scaling BFGS method
global convergence
