一种由滤子松弛化的序列二次规划算法

A Kind of SQP Methods Relaxed by Filter

提出了一种新的线搜索罚函数方法。它将滤子方法与其相结合,事实上是滤子方法的一种应用。与以前的滤子方法不同,此法不需要可行性恢复阶段,在一定的假设条件下,它可以得到全局收敛性;通过二阶校正,还可以得到局部超线性收敛性。数值结果表明该方法是非常有效的。

This paper, introduces a new algorithm for nonlinear optimization which applies filter techniques to the traditional line search penalty function method. Unlike formal filter methods, do not need restoration phase here. And under reasonable assumptions, global convergence is given. Furthermore, by second order correction, local convergence is proved. Numerical results are presented which show the robustness of the algorithm.