一种基于向量的鲁棒ARMA系统辨识器

A Robust ARMA System Identification Based on the Vector Method

提出一种新的基于向量方法的自回归和运动平均(ARMA)模型系统辨识器,并给出了其参数的统计分析模型。应用结果表明,向量ARMA算法和最小二乘法LS算法相比,在一定条件下,其预测误差精度提高了约1 2dB;且该系统模型不受分离向量参数的影响。使用非线性函数核,系统将会成为一个鲁棒的非线性辨识过程。

To present a new approach to auto-regressive and moving average(ARMA) modeling based on the support vector method,a statistical analysis of the characteristics of the proposed method is carried out.The results show, compared SVM-ARMA with LS,precisions of validation prediction error of the SVM-ARMA improved 1.2 dB than LS in some conditions. Besides, the effect of outliers can be cancelled.With using nonlinear kernels,the system will lead to robust, nonlinear system identification procedures.