1、第39卷第4期2023年8月山西大同大学学报(自然科学版)Journal of Shanxi Datong University(Natural Science Edition)Vol.39 No.4Aug.2023一种基于L-函数的非单调自适应信赖域算法张杰,朱子旋,芮绍平,曾柔(淮北师范大学 数学科学学院,安徽 淮北 235000)摘要:利用函数L-就无约束优化问题提出了一种非单调自适应信赖域算法。算法中信赖域半径自动更新依赖函数L-,步长的求解采用了非单调wolfe线搜索技术。在一定条件下,证明了算法的全局收敛性,数值实验表明算法稳定有效。关键词:无约束优化;信赖域算法;自适应策略;全局
2、收敛性中图分类号:O224文献标识码:Adoi:10.3969/j.issn.1674-0874.2023.04.005考虑无约束最优化问题1minx Rnf(x),(1)其中:f:Rn R为连续可微函数。信赖域方法是求解无约束优化问题的一类重要的数值方法,由Powell2提出,该方法具有良好的适应性与稳定性。对于无约束优化问题,传统的信赖域方法利用二次函数逼近函数f,得到信赖域算法的子问题3:mink(d)=gTkd+12dTBkd,s.t.d2 k,(2)其中:gk=f(xk);Bk为Hesse矩阵2f(xk)的第k次近似;k为信赖域半径;是Euclidean范数。引入Aredk=f(xk
3、)-f(xk+dk)和Predk=k(0)-k(dk)分别代表目标函数的实际下降量和预测下降量4。令rk=AredkPredk。传统的信赖域方法中,信赖域半径k在给定的比率下调整,如果rk 2,L(t)1,(7)limt +L(t)=3,这 里 的 常 数 满 足:0 c1 c2 2 3 1 2-1。由文献8可知L-函数有界。采用的信赖域半径更新规则为k+1=L(rk)k,其中L(rk)是一个关于比值rk的L-函数,结合非单调技术将比率rk更改为:rk=Dk-f(xk+dk)k(0)-k(dk),基于此更新策略,设计算法1。算法1 非单调自适应线搜索信赖域算法(NTTR)步 0 给定x0 Rn
4、,B0 Rn n和0 0,(0,1),0 c1 c2 1,0 2 3 1 1,0 1 1,(0,1),0 0,M 1,令k=0;步 1 计算gk=f(xk),如果gk,则算法终止;否则转步2;步 2 求 解 信 赖 域 子 问 题(2)的 近 似 解dk,dk k;步3 计算rk=Dk-f(xk+dk)k(0)-k(dk);步 4 若rk,令xk+1=xk+dk,否则求步长k,满足非单调wolfe线搜索:f(xk+kdk)Dk+kgTkdk,g(xk+kdk)Tdk gTkdk。(5)令xk+1=xk+kdk;步5 更新信赖域半径:k+1=L(rk)k;步6 更新Bk,令k=k+1,转步1。2
5、 收敛性分析为了证明算法的收敛性,以下假设条件成立:A1:水平集L(x0)=x Rn|f(x)f(x0)。有界且f(x)在水平集上连续可微有下界;A2:g(x)=f(x)在水平集上是Lipschitz连续,即存在常数L 0,使x,y Rn,f(x)-f(y)Lx-y;A3:序列 Bk一致有界,即M1 0使得k N,有Bk 0,使得gk。下证limk dk=0。由算法1,产生的迭代点列满足:f(xl(k)-f(xk+dk)k(0)-k(dk)Dk-f(xk+dk)k(0)-k(dk),即f(xl(k)-f(xk+dk)(k(0)-k(dk)0。于是fl(k)-1-fl(k)(k(xl(k)-1)
6、-k(xl(k)0。(6)又 由m(k+1)m(k)+1及fl(k)定 义 知:fl(k+1)fl(k),即fl(k)单调递减,再由假设条件 A2和A3知有其有下界,所以fl(k)收敛。因此对(6)式两边取极限可得:limk k(xl(k)-1)-k(xl(k)=0。设s 0,使得dk满足:dk sgk。由引理 1可知:k(xl(k)-1)-k(xl(k)gl(k)-1minl(k)-1,gl(k)-1Bl(k)-1smin1,1sM1dl(k)-12 0。因此limk dl(k)-1=0。根据f(x)的连续性,可得limk fk+1=limk fl(k)。于是fl(k)-f(xk+1)-k(
7、dk),两边同时取极限得limk k(dk)=0,即limk dk=0。262023年又据引理2,引理3有:|rk-1(dk2)k(0)-k(dk)(k2)min k,M1 0。对于k N,当k +有rk=1,结合引理 1得:k(dk)gkmink,gkBk 0,矛盾,故假设不成立,定理得证。3 数值实验给出基于L-函数的非单调自适应信赖域算法(NTTR)的数值结果,并且和传统的信赖域算法(TTR)进 行 比 较。算 法 运 行 环 境 是 MATLAB(R2022a)。参数选取如下:0=g(x0),B0=I。终止准则为g(xk)10-8。算法中信赖域子问题(2)采用折线法求解,信赖域半径调节
8、函数采用文献8中的L-函数:L(rk)=c1+(c2-c1)exp(rk),rk 0,-(1-2)exp(1)1-exp(1)exp(rk-1)+1-2exp(1)1-exp(1),0 rk 1,1,1 rk 2-1其中:1=2.0,2=0.5,3=0.7,=0.25,=0.4,=0.5,M=10,1=0.75,c1=0.12,c2=0.14,=0.1,=0.3,min=0.13,max=0.89。计算结果见表1和表2,其中dim代表测试的维数,num代表算法迭代次数,val代表最优值。算例1和算例2来自文献15。算例1 测试函数为:f(x)=c(x2-x21)2+(1-x1)2,c=100。
9、算例2 测试函数为:f(x)=i=1n 4(x4i-1-10 x4i-2)2+5(x4i-1-x4i)2+(x4i-2-2x4i-1)2+10(x4i-3-10 x4i)4。初始点x0=(3,-1,0,3,.3,-1,0,3)T。表1 算例1数值实验结果初始点(x0)(1,0,-1,0)T(-1,0,1,0)T(2,0,2,0)TnumTTR162115NTTR112015valTTR1.66 10-162.11 10-196.55 10-22NTTR4.66 10-222.21 10-202.95 10-22表2 算例2数值实验结果dim100500100 0500 0numTTR83958
10、6103NTTR77756988valTTR3.96 10-123.88 10-111.87 10-124.28 10-11NTTR2.69 10-145.82 10-142.08 10-134.19 10-12从表1可以看出,对于不同的初始点选择,由于非单调技术的引入,降低了算法的计算量,迭代次数小于传统的信赖域算法。从表2可以看出,随着测试问题从低维到高维,新的非单调自适应信赖域算法能快速有效地得到最优解,并且具有良好的稳定性,说明算法NTTR适合求解较大规模问题。4 结语信赖域方法是求解无约束优化问题的经典方法之一,对该方法改进具有重要的实际意义和应用价值。基于L-函数的非单调自适应信赖
11、域算法很好地解决传统信赖域方法中信赖域半径更新问题,实现了自动更新迭代,非单调技术的使用也大大减少了算法的计算量,调高了算法效率,因此这种方法丰富和推广了现有结果。参考文献1 NOCEDAL J,WRIGHT S J.Numerical optimization M.New York:Springer,2006.2 POWELL M J D.On the global convergence of trust region algorithms for unconstrained minimization J.Math Program 1984,29(3):297-303.3 RAMIREZ
12、V A,Sottosanto G N.Nonmonotone trust region algorithm for solving the unconstrained multiobjective optimizationproblems J.Comput Optim Appl,2022,81(3):769-788.4 SANG Z,SUN Q.A self-adaptive trust region method with line search based on a simple subproblem model J.J Comput ApplMath,2009,232(2):514-52
13、2.5 KAMANDI A,AMINI K,AHOOKHOSH M.An improved adaptive trust-region algorithm J.Optim Lett,2017,11(3):张杰等:一种基于L-函数的非单调自适应信赖域算法27山西大同大学学报(自然科学版)2023年555-569.6 CUI Z,WU B.A new self-adaptive trust region method for unconstrained optimization J.J Vib Control,2012,18(9):1303-1309.7 HEI L.A self-adaptive
14、 trust region algorithm J.J Comput Math,2003,21(2):229-236.8 LU Y,LI W,CAO M,et al.A novel self-adaptive trust region algorithm for unconstrained optimization J.J Comput Appl Math,2014,2014(1):1-8.9 LARSON J,MENICKELLY M,WILD S M.Derivative-free optimization methods J.Acta Numer,2019,28:287-404.10 G
15、U N,MO J.Incorporating nonmonotone strategies into the trust region method for unconstrained optimization J.ComputMath appl 2008,55(9):2158-2172.11 AHOOKHOSH M,AMINI K.An efficient nonmonotone trust-region method for unconstrained optimization J.Numer Algo-rithms,2012,59(4):523-540.12 李小伟,钱慧敏.带有线搜索的
16、非单调自适应新锥模型信赖域算法 J.电子科技,2013,26(11):4-6.13 YAN X,HE Q,WANG Y.Truncated trust region method for nonlinear inverse problems and application in full-waveform inver-sion J.J Comput appl Math,2022,404:1-12.14 ESMAEILI H,KIMIAEI M.An efficient adaptive trust-region method for systems of nonlinear equations
17、 J.Int J ComputMath,2015,92(1):151-166.15 TOUATIAHMED D,STOREY C.Efficient hybrid conjugate gradient techniques J.J Optimiz Theory App,1990,64(2):379-397.ANon-monotoneAdaptiveTrust RegionAlgorithm Based onL-functionZHANG Jie,ZHU Zi-xuan,RUI Shao-ping,ZENG Rou(School of Mathematical Science,Huaibei N
18、ormal University,Huaibei Anhui,235000)Abstract:In this paper,we propose a non-monotonic adaptive trust region algorithm for unconstrained optimization problemsusingL-function.In the algorithm,the automatic update of the radius of the trust region depends on theL-function,and thestep size is solved b
19、y using the non-monotonic Wolfe line search technique.Under certain conditions,the global convergence ofthe algorithm is proved.Numerical experiments show that the algorithm is stable and effective.Key words:unconstrained optimization;trust region algorithm;self-adaptive method;global convergence责任编
20、辑 高彩云(上接第16页)It is one of the core technologies in the field of modern wireless communication technology and wireless broadband mobile,andhas been widely used.Based on the difference in the number of antennas at the receiving and transmitting ends,MATLAB soft-ware is used to study the channel capaci
21、ty of SISO,SIMO,MISO and MIMO channels in AB MIMO system,compare the chan-nel capacity of channel state information(CSI)in AB MIMO system when the sender is known or unknown,and verify the supe-riority of the MIMO system.Key words:multiple-input multiple-output;water injection algorithm;wireless communication;average power allocation algo-rithm;ergodic capacity责任编辑 高彩云28