结合锥模型算法的修正多步拟牛顿法

Modified Multi-step Quasi-Newton Methods Combining Conic Model Algorithm

提出一种解决维数较大的无约束优化问题的混合算法.该算法采用了一个开关用于切换修正多步拟牛顿算法和锥模型算法.结果表明,该算法既保留了拟牛顿算法的快速有效性,又将适用范围扩大到了二次模型逼近效果较差的函数,且对于大多数维数较大的函数都是切实可行的.

We introduce a hybrid method for solving middle-large dimension unconstrained optimization. The method adopts a on-off which switch modified multi-step quasi-Newton methods and conic model algorithm. The new hybrid method keeps the efficiency of quasi-Newton method as well as to be broadly applied to the functions which can not get accurate result based on the standard methods. Our new method is feasible to most middle-large dimension function.