模拟退火法及其在抗差估计中的应用

Simulaed anneling method and its application in the robust estimation

抗差估计结果的好坏与初值的选取密切相关。当迭代初值偏离较大时,ρ函数不是严凸时,采取的选权迭代法往往不能获得全局最优值,而只能得到局部的最优解。本文提出一种新的抗差估计求解过程,即把它看作是关于未知参数的全局最优化问题,应用模拟退火法进行求解。计算结果表明:相对于传统的选权迭代法,模拟退火法对初值的选取不敏感,能给出全局最优值。

The result of robust estimation is quite related to the initial values. When the initial value lies far away from the real solution andpfunction is not a strictly convex function, the usual strategy of iteratively reweighed least squares will not get the right solution, probably it only leads to the local optimization. In this paper, the authors propose a method to solve robust estimation as a global optimization problem based on simulated annealing method. Result shows that the simulated annealing method is insensitive to initial value and it can get the right solution in contrast with traditional strategy.