XIONG Qian,WU Zezhong.Research on Generalized Lagrange Multiplier Method based on BFGS Algorithm[J].Journal of Chengdu University of Information Technology,2020,35(02):221-234.[doi:10.16836/j.cnki.jcuit.2020.02.014]
基于BFGS算法的广义Lagrange乘子法研究
- Title:
- Research on Generalized Lagrange Multiplier Method based on BFGS Algorithm
- 文章编号:
- 2096-1618(2020)02-0221-14
- Keywords:
- applied mathematics; optimization theory; constrained optimization; generalized Lagrange multiplier method; penalty factor; correction factor
- 分类号:
- O224
- 文献标志码:
- A
- 摘要:
- 广义Lagrange乘子法是解决约束优化的问题的一种重要方法,基于BFGS算法,利用MATLAB工具,研究了初始点的选取、罚因子的改变及罚因子修正系数的改变对该算法收敛效果的影响。结果表明:(1)对于初始点的选取,应尽量在最优点附近进行取值,才能有不错的收敛效果。(2)罚因子过小或过大都对算法求解问题产生困难。如果罚因子太小,大量的搜索时间将花费在非可行域,使迭代次数增加。另一方面,如果罚因子过大,算法将很难被推进到可行域以内,导致算法收敛失败。(3)随着罚因子修正系数的变化,随时会出现无法收敛的现象,故该系数的值应在迭代成功率相对较高的分段选取。
- Abstract:
- The generalized Lagrange multiplier method is an important method to solve the problem of constrained optimization. Based on BFGS algorithm, this paper uses MATLAB tool to study the influnence of initial point selection, penalty factor change and penalty factor change correction coefficient. The results show that:(1)For the selection of the initial point, the value should be selceted near the best point, and the convergence effect can be good.(2)The penalty factor too small or too large will make the algorithm difficult to solve. If the penalty factor is too small, a large amount of search time will cost in the non-feasible domain, which will increases the number of iterations. On the other hand, if the penalty factor is too large, the algorithm will be difficult to be pushed into the feasible domain, which will causes the algorithm fail to converge.(3)With the change of the penalty factor correction coefficient, there is a phenomenon of fail to converge at any time. Therefore, the value of the coefficient is supposed to selct during the segment of iteration success rate that is relatively high.
参考文献/References:
[1] Hestenes M R.Multiplier and gradient methods[J].Journal of Optimization Theory and Applications,1969,4(5):303-320.
[2] Rockafellar R T.The multiplier method of Hestenes and Powell applied to convex programming[J].Journal of Optimization Theory and Applications,1973,12(6):555-562.
[3] Rockafellar R T.A dual approach to solving nonlinear programming problems by unconstrained optimization[J].Mathematical Programming,1973,5(1):354-373.
[4] Davidon W C.Optimally conditioned optimization algorithms without line searches[J].Mathematical Programming,1975,9(1):1-30.
[5] Broyden C G.The convergence of a class of double-rank minimization algorithms[J].J.Institute of Mathematics and Its Applications,1970,6(1):76-90.
[6] Williams T M.Practical Methods of Optimization.Unconstrained Optimization[J].Journal of the Operational Research Society,1981,32(5):417.
[7] 刘兴高,胡云卿.应用最优化方法及MATLAB实现[M].北京:科学出版社,2014.
[8] Fiacco A V,McCormick,Garth P.Nonlinear Programming:Sequential Unconstrained Minimization Techniques[J].Philadelphia:SIAM,1990.
[9] 宋菲,吴泽忠.外罚函数法与广义Lagrange乘子法的比较研究[J].成都信息工程大学学报,2017(6).
[10] Andreani R,Martínez,J.M,Santos L T,et al.On the behaviour of constrained optimization methods when Lagrange multipliers do not exist[J].Optimization Methods and Software,2014,29(3):646-657.
[11] 陈宝林.最优化理论与算法[M].2版.北京:清华大学出版社,2005.
[12] 袁亚湘,孙文瑜.最优化理论与方法[M].上海:科学出版社,2001.
相似文献/References:
[1]胡春华.基于分数阶导数和变分迭代法的人口预测算法[J].成都信息工程大学学报,2017,(01):78.[doi:10.16836/j.cnki.jcuit.2017.01.013]
HU Chun-hua.An Algorithm of Population Forecast with Fractional
Derivative and Variational Iteration Method[J].Journal of Chengdu University of Information Technology,2017,(02):78.[doi:10.16836/j.cnki.jcuit.2017.01.013]
[2]卞广钱,周 磊.基于模糊贴近度的属性约简[J].成都信息工程大学学报,2017,(01):86.[doi:10.16836/j.cnki.jcuit.2017.01.015]
BIAN Guang-qian,ZHOU Lei.Attribute Reduction based on Fuzzy Closeness Degree[J].Journal of Chengdu University of Information Technology,2017,(02):86.[doi:10.16836/j.cnki.jcuit.2017.01.015]
[3]王 容,罗文力,廖群英.方程φ3(n)=n/d 的可解性[J].成都信息工程大学学报,2017,(01):95.[doi:10.16836/j.cnki.jcuit.2017.01.017]
WANG Rong,LUO Wen-li,LIAO Qun-ying.On the Solvability of the Equation φ3(n)=n/d[J].Journal of Chengdu University of Information Technology,2017,(02):95.[doi:10.16836/j.cnki.jcuit.2017.01.017]
[4]马 斌,吴泽忠.基于人工蜂群算法的供应链网络均衡问题研究[J].成都信息工程大学学报,2017,(03):336.[doi:10.16836/j.cnki.jcuit.2017.03.016]
MA Bin,WU Ze-zhong.Research on Supply Chain Network Equilibrium Model
based on Artificial Bee Colony Algorithm[J].Journal of Chengdu University of Information Technology,2017,(02):336.[doi:10.16836/j.cnki.jcuit.2017.03.016]
[5]丁云红,陈勇明.基于改进灰靶决策的女子七项全能排名模型[J].成都信息工程大学学报,2017,(06):678.[doi:10.16836/j.cnki.jcuit.2017.06.018]
DING Yun-hong,CHEN Yong-ming.Women's Heptathlon Ranking Model based
on Improved Gray Target Decision[J].Journal of Chengdu University of Information Technology,2017,(02):678.[doi:10.16836/j.cnki.jcuit.2017.06.018]
[6]黄 艳,吴泽忠.基于Lévy飞行的一种改进鲸鱼算法[J].成都信息工程大学学报,2021,36(01):24.[doi:10.16836/j.cnki.jcuit.2021.01.005]
HUANG Yan,WU Zezhong.An Improved Whale Algorithm based on Lévy Flight[J].Journal of Chengdu University of Information Technology,2021,36(02):24.[doi:10.16836/j.cnki.jcuit.2021.01.005]
[7]黄 飞,吴泽忠.基于Armijo搜索步长的几种共轭梯度法的分析对比[J].成都信息工程大学学报,2019,(02):209.[doi:10.16836/j.cnki.jcuit.2019.02.0017]
HUANG Fei,WU Zezhong.Analysis and Comparison of Several Conjugate Gradient
Methods based on Armijo Search Step Length[J].Journal of Chengdu University of Information Technology,2019,(02):209.[doi:10.16836/j.cnki.jcuit.2019.02.0017]
备注/Memo
收稿日期:2019-08-12 基金项目:国家自然科学基金资助项目(71672013)