解非线性规划问题的不精确线性搜索SQP滤子方法(英文)  

An Inexact Line Search SQP-Filter Method for Nonlinear Programming Problems

作　　者：刘美玲[1,2] 濮定国[1] 李学迁[3] 

机构地区：[1]同济大学数学系,上海200092 [2]南昌工程学院理学系,江西南昌330099 [3]上海大学管理学院,上海200444

出　　处：《应用数学》2011年第3期532-539,共8页Mathematica Applicata

基　　金：Supported by the National Science Foundation of China(10771162)

摘　　要：本文用序列二次规划方法(SQP)结合Wolfe-Powell不精确线性搜索准则求解非线性规划问题.Wolfe-Powell准则是一种能够使目标函数获得充分下降而运行时间较省的确定步长方法.不精确线性搜索滤子方法比较其它结合精确线性搜索和信赖域方法求解问题的滤子方法更灵活更易实现.如果目标函数的预测下降量为负,我们的工作将主要利用可行恢复项改善可行性.一般条件下,本文提出的算法较易实现,且具有全局收敛性.数值试验显示了算法的有效性.In this paper,we solve nonlinear programming problem by a sequential quadratic programming（SQP） algorithm combined with Wolfe-Powell inexact line search criterion. Wolfe-Powell criterion is a more practical strategies to identify step length that achieves adequate reductions in objective function at minimal cost. And the inexact line search filter method is more flexible and realizable comparing with exact one and trust region methods applied in most problems solved by filter method. When the predict reduction of QP-subproblems is not positive,we turn mainly our work to improve feasibility by a so-called feasibility restoration phase. Under mild conditions,our SQP-filter method is more easier to realize and can convergent globally. Numerical results of test problems show efficiency of this method.

关 键 词：非线性规划 Wolfe-Powell准则 SQP 滤子 收敛性 

分 类 号：O224[理学—运筹学与控制论]