一般约束优化问题的一个新广义梯度投影法

A New Generalized Gradient Projection Method for General Constrained Optimization

本文以l1-l∞混合罚函数作为效益函数,提出一个一般约束优化问题的新的广义梯度投影法.该算法具有以下特点:初始点可任意选取;搜索方向是效益函数的下降方向;在传统的假设条件下具有全局收敛性.论文最后通过数值试验验证了算法的有效性.

In this paper,we propose a new generalizd gradient projection method for gen eral constrained optimization based on a hybrid l1 -- l∞ penalty function. The characters of the proposed algorithm are as follows.the start point can be chosen arbitrarily;the searching di- rection is a decreasing direction of the merit function;and the algorithm is global convergent under the traditional assumptions. Finally,some preliminary numerical results show that the proposed algorithm is effective.