一种基于模拟退火思想的锥模型信赖域方法

A Conic Model Trust Region Method Based on Idea of Simulated Annealing

针对无约束优化问题,提出一种基于简化锥模型的信赖域方法.该方法将模拟退火思想改进的Metropolis准则和简化锥模型信赖域方法相结合,即用标量矩阵近似Hessian矩阵,再由Metropolis准则接受更多的试探步.在适当的条件下,建立了新算法的全局收敛性.数值结果表明,该方法是有效的.

A trust region method based on simple conic model is proposed for unconstrained optimization prob-lems.The Metropolis criterion improved by idea of simulated annealing is combined with the simple conic model trust region method,namely,Hessian matrix is approximated by a scalar matrix,and the Metropolis criterion is used to accept more trial steps.Under appropriate conditions,the global convergence of algorithm is established.Numerical results show that the new method is effective.