基于信赖域子问题的共轭梯度法

Trust Region Based on Conjugate Gradient Method

对于求解无约束优化问题,利用重新开始的三项共轭梯度法与信赖域方法结合,并引入非单调技术,当迭代不成功时,改进后的算法保留一些有用信息,提高了算法的有效性。在适当的条件下,给出了新算法的全局收敛性,数值试验结果表明新算法是有效的。

For solving unconstrained optimization problems, the three restart conjugate gradient method was combined with trust region methods and the non-monotonic techniques was introduced. When the iteration is unsuccessful, the improved algorithm was to keep some useful information to improve the effectiveness of the algorithm. Under the right conditions, the global convergence of the new algorithm was given, and the numerical results show that the new algorithm is effective.