一种多重滤子非单调的新锥模型信赖域算法

非单调技术在信赖域中得到广泛应用,并取得了较好的结果,但传统的非单调技术存在遗漏最优点等缺陷,文中提出了一种新的基于新锥模型的非单调多重过滤信赖域方法,该算法不但在每次迭代中构造出新的比率来调整信赖域半径减少运算量,同时在实验步骤不被接受时利用多重过滤技术增加了试验点的接受几率,并在适当的条件下,证明了算法的收同时敛性。数值试验则表明了该算法的有效性。

A New Nonmonotone Multidimensional Filter Trust Region of a New Conic Model

Nonmonotone technique has been widely applied in the trust region with good results. A new filter trust region method is proposed by combining nonmonotone techniques with the multidimensional filter technology. The algorithm constructs a new ratio in each iteration to adjust trust region radius of the computation, and uses multi- ple filtering to improve the probability of acceptance and the experiment when the test steps are not accepted. The global convergence of algorithm is proved under certain conditions. Numerical experiments show that the algorithm is effective.