CAJ | 학술논문

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.
본문용서렬이차규화방법(SQP)결합Wolfe-Powell불정학선성수색준칙구해비선성규화문제.Wolfe-Powell준칙시일충능구사목표함수획득충분하강이운행시간교성적학정보장방법.불정학선성수색려자방법비교기타결합정학선성수색화신뢰역방법구해문제적려자방법경령활경역실현.여과목표함수적예측하강량위부,아문적공작장주요이용가행회복항개선가행성.일반조건하,본문제출적산법교역실현,차구유전국수렴성.수치시험현시료산법적유효성.
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.