非线性回归的自动化建模过程

Nonlinear Regression Automatic Modeling Process

提出一种非线性回归的自动化建模过程。文中将基于Gram-Schmidt过程的回归方法与交叉有效性分析相结合,首先构造出备选的回归模型集合,然后通过进行多次交叉有效性验证的方法,对构造出的一系列的备选模型,采用投票方式,挑选出被选中次数最多的模型作为最终确定的回归模型。该建模过程形成一种自动确定非线性回归模型的机制。仿真研究表明,采用本文所提出的自动化建模方法,可以合理有效地确定最终模型,并且模型具有良好的稳健性和预测效果。

A nonlinear regression automatic modeling process was proposed in this paper. First, a set of candidate regression models were constructed based on Gram-sehmidt process with cross-validation analysis; then, cross-validation method was repeatedly used to find the models for votes. The model getting most votes was selected as the final regression model. Therefore, an automatic determining mechanism of nonlinear regression model can be obtained through modeling process. Simulation indicates the proposed automatic modeling method proposed can determine the final model rationally and effectively, and the selected model possesses a good robustness and predictive ability.