逐步回归法辨识线性系统模型

线性系统的结构辨识主要依据模型残差来人工判断。这不仅引入了判断误差,同时也使该方法难于投入实际应用。针对这一问题,该文提出一种修正的逐步回归法,用于对有色噪声环境下的线性系统建模。该方法将各回归变量按对输出的“贡献”排队,通过判断“贡献”的显著性来决定是否选入回归方程,从而能够同时决定模型的阶次、系统纯时延和模型参数。该方法具有计算量小、适应性强等优点。仿真试验验证上述结论。

Identification of Linear System Model with Stepwise Regression

The method of structure identification for linear system is still determined by users according to the residual of model.It is not only difficult to use the method in reality,but also to make the error judgement imported.To solve the problem,this paper presents a fixed stepwise regression modeling method for linear system with color noise.A "contribution" of regression variable to output is introduced and listed from the high to the low.Then the variable is introduced to regression model if the "contribution" is remarkable.It detects model order and time delay while estimating the corresponding parameters,having less compution,more powerful adaptability,and son on.A simulation test is given to verify the results proposed.