一种无约束优化问题的谱共轭梯度法  

A Nomonotone Spectral Conjugate Gradient Methods for Unconstrained Optimization

作　　者：朱花[1] 王希云[1] 

机构地区：[1]太原科技大学应用科学学院,太原030024

出　　处：《太原科技大学学报》2010年第3期245-249,共5页Journal of Taiyuan University of Science and Technology

基　　金：山西省自然科学基金(2008011013)

摘　　要：提出了一种新的谱共轭梯度法,证明了该方法不依赖于任何线搜索具有充分下降性,在Armijo线搜索下证明了算法具有全局收敛性。数值试验结果表明:在Armijo线搜索下,该方法比Necu-lai,Andrei提出的方法有效;并且4种测试函数的数值结果显示:新方法明显优于谱DY算法,也较谱FR算法有效;可以和谱PRP的计算效能相媲美,故算法具有良好的计算效能。A new nonmonotone spectral conjugate gradient method line search technique is proposed in this paper.The global convergence of corresponding algorithm with the proposed line search is proved.Preliminary numerical results show that this method is very efficient and suitable for solving large-scale unconstrained optimization.Under the Armijo line search,this method more efficient than Neculai Andersi＇s a conjugate gradient formula.The numerical results of four test funtions show that new method is better than DY spectrum algorithm,and effective than FR method,which is comparable to the calculation of performance PRP spectrum.So the algorithm has good performance computing.

关 键 词：无约束优化 谱共轭梯度法 线搜索 全局收敛性 

分 类 号：O221[理学—运筹学与控制论]