LXy1ii2、y2II2b
(3)的共轭方法是全局收敛的。
根据引理3可知算法1具有全局收敛性。
3数值实验
本文算法的数值实验结果如表1,算法的参
数5 = 0_01,£7 = 0.1,= 10—5,终止条件为|丨心|丨矣 心或迭代次数不超过1〇4。
表
1
数值实验
ProblemDimnewMPRPNI/NF/NGNI/NF/NGJENSAM215/85/2611/30/20BARD320/44/3117/41/29GAUSS34/9/54/9/5GULF31/2/21/2/2BOX31/51/21/51/2KOWOSB4563/2236/955563/2236/955
0SB1
51/51/21/51/2BIGGS6109/436/1798/260/145SINGX414/176/9414/176/94PEN125/18/125/18/12PEN2420/96/3617/140/34VARDIM23/9/73/9/7VARDIM5010/52/3610/52/36TRIG
3106/638/178106/638/178TRIG50209/2063/301209/2113/302
BV1013/27/1713/27/17IE35/12/75/12/7IE
50
6/13/7
6/13/7
26
续表
钦州学院学报 第33卷
由表可知,算法1基本保持参数为氏_的算
ProblemIETRIDTRIDTRIDBANDLINLINLINLINLIN1LIN1LINO
Dim1003100200325050010002104
newNI/NF/NG6/13/814/36/1931/69/3731/162/367//121/3/31/3/31/3/31/3/31/51/21/3/31/3/3
MPRPNI/NF/NG6/13/818/86/2431/69/3731/162/367/63/111/3/31/3/31/3/31/3/31/51/21/3/31/3/3
法,是有效的。相对于参数为汍_的算法,算法1 的一个特点是&〇,另一个特点是搜 索方向在强《w/e线搜索下保持的充分下降性的 参数〇■范围较小。
参考文献
[1 ]戴或虹,袁亚湘.非线性共轭梯度法[M ].上海:上海科学技 术出版社,2001.
[2] WEI Z, YAO S,LIU L. The convergence properties of some con
表中PraWem表亦测试冋题的名称;Dim表;^ 目标函数变量的维数;表示算法1
表
示与强线搜索一起的算法;yv//yvF/yvG 分别表示迭代次数/目标函数值计算次数/目标函 数梯度值计算次数。
jugate gradient methods[ J]. Applied Mathematics and Computa
tion, 2006 (183) :1341-1350.[3] HUANG H D,LI Y J,WEI Z X. Global convergence of a modi
fied PRR conjugate gradient method [ J ]. Journal of Mathematical Research & Exposition,2010(30) : 141-148.[4] Wolfe P. Convergence conditions for ascent methods. SIAM Re
view, 1969( 11) :226-235.[5] Wolfe P. Convergence conditions for ascent methods. II: Some
correction,SIAM Review,1971(31) :185-188.[6] Zoutendijk G. Nonlinear Programning( Abadie J,ed. ) [ M]. Am
sterdam :North-Holland, 1970.[7] Z F Li, J. Chen,N Y Deng, convergence properties gradient meth
od with Goldstein line search. J. China Agric. Univer, 1996(4):
15-18.
[8 ] Glibert J C, Nocedal J. Global convergence properties of conju
gate gradient methods for optimization [ J ] . SIAM. J. Optimization, 1992(2) :21-42.
A New Conjugate Gradient Method of Global Convergence
under the Strong wolfe Line Search
Abstract : Base on the improved PRP conjugate gradient method, a new conjugate gradient method is presented for uncon
strained optimization problems. The algorithm possesses sufficient descent property and global convergence under the strong wolfe line search. The numerical results show that the algorithm is effective.
Key words : unconstrained optimization ; conjugate gradient method ; sufficient descent property ; global convergence ; numerical experiment
「责任编辑黄立壮]
LI Zhiqun,LU Jing,ZHANG Shuang
(College of Science, Qinzhou University, Qinzhou 535011 , China)