基于模拟退火算法的最小一乘回归新算法

A New Algorithm for the Least Absolute Deviation Regression Based on the Simulated Annealing Algorithm

最小一乘准则由于其稳健性较好而在工程中得到广泛的应用,但求解最小一乘回归模型系数的算法往往过于复杂或只能用于样本和变量个数较少的情形.本文根据最小一乘的性质,把最小一乘问题变为组合优化问题,将模拟退火算法用在最小一乘模型的求解上,在后面的数值实验中取得了较好的效果。

The least absolute deviation criteria is widely used in engineering because of its robustness, but the algorithms for solving the least absolute deviation estimate of the regression coefficient are too explicated or only efficient for small sample points and variables. In this paper, a new method based on simulated annealing algorithm to solve the least absolute deviation estimates of regression coefficient is presented by changing the problem to the combinatorial optimization based on it＇s properties. At last the numerical experimentations verified the validity of the new method.