改进的次梯度优化算法的收敛性研究  

Research of a Modified Subgradient Optimization Algorithm' Astringency

作　　者：陈静[1] 于伟伟 倪明放[3] 

机构地区：[1]金陵科技学院公共基础课部,江苏南京211169 [2]61600部队北京83信箱,北京100000 [3]中国人民解放军理工大学通信工程学院,江苏南京210007

出　　处：《金陵科技学院学报》2008年第3期12-16,共5页Journal of Jinling Institute of Technology

基　　金：江苏省高校自然科学基金(06KJD110068)

摘　　要：次梯度优化算法是求解Lagrange对偶问题的一种有效的算法。不同的步长策略对算法的收敛性有不同的影响,I型锯齿现象对算法的收敛速度也会产生影响。通过改进步长因子公式和搜索方向,在普通次梯度优化算法的基础上,提出了一种改进的次梯度优化算法(MSOA)。该算法具有收敛、能够有效地消去I型锯齿影响的优点。The Subgradient Optimization Algorithm is used to solve Lagrange Dual problem. The convergence properties of Subgradient Optimization Algorithm is affected by different step length strategy and the usually slowly convergence which is mainly caused by some zigzagging phenomena. Then, the Modified Subgradient Optimization Algorithm （MSOA）, which has the proper step length strategy and proper step direction strategy based on General Subgradient Optimization Algorithm （GSOA）, is presented. MSOA has the advantage of convergence and will effectively eliminate the influence of some zigzagging phenomena.

关 键 词：步长策略 搜索方向 次梯度优化算法 Ⅰ型锯齿 

分 类 号：O221.2[理学—运筹学与控制论]