新锥模型自适应过滤算法

A Filter Self-adaptive Algorithm Based on the New Conic Model

提出了一种新的过滤算法.在新锥模型信赖域的算法中引入过滤技术,这种算法加大了试验点被接受的几率,且避免了原算法中子问题重复计算的问题.在适当的条件下,证明了这种算法的全局收敛性.针对几个具体的函数进行了数值实验.

This paper presents a novel filter algorithm. The new cone model trust region algorithm introduces filtering technique,which increases the probability of test points to be accepted and avoids the double counting problem of sub-problems in the original algorithm. Under appropriate conditions,the global convergence of this algorithm is proved. Numerical experiments on several specific functions are carried out.