基于新锥模型的带固定步长的非单调自适应信赖域算法  

A nonmonotone self-adaptive trust region algorithm with fixed stepsize based on the new conic model

作　　者：朱帅[1] 朱世昕[1] 王希云[2] 

机构地区：[1]山西大同大学,山西大同037003 [2]太原科技大学应用科学学院,山西太原030024

出　　处：《西南民族大学学报（自然科学版）》2012年第1期44-49,共6页Journal of Southwest Minzu University(Natural Science Edition)

基　　金：山西大同大学校级青年项目(2007Q14)

摘　　要：对于无约束优化问题,在传统信赖域算法中通常采用二次模型来逼近函数,而对某些非二次性态较强,曲率变化剧烈的函数,用该方法逼近效果较差,于是本文提出一类基于新锥模型带固定步长的非单调自适应信赖域算法.利用一种固定步长算法,在一定条件下证明了该算法的全局收敛性及收敛速率.数值实验表明该算法是有效的.For unconstrained optimization,quadratic model function is often used in the traditional trust region algorithm to approximate function.but for some functions which show strong none-quadric features and dramatic changes in curvature,the effect is poor.This paper puts forward a new nonmonotone adaptive trust region algorithm model based on a cone with constant step length.According to a fixed step length algorithm in some conditions,the algorithm is proved to be global convergence and fast convergence rate.The numerical experiments show that this algorithm is effective.

关 键 词：无约束优化 非单调技术 自适应信赖域算法 固定步长 新锥模型 

分 类 号：O221.2[理学—运筹学与控制论]