非线性系统定阶的神经网络方法

System order determination using radial basis function neural networks

前馈型神经网应用于非线性系统辨识的一个问题是确定系统阶次。采用前馈神经网进行非线性系统定阶与神经网的推广性问题密切相关。ＯＬＳ算法是构筑径向基神经网的一种学习算法，但是采用ＯＬＳ算法构筑神经网存在推广性问题。ＲＯＬＳ算法将ＯＬＳ算法与正则化（ｒｅｇｕｌａｒｉｚａｔｉｏｎ）方法相结合，以提高算法的推广能力。本文将基于径向基网的ＲＯＬＳ算法应用于非线性系统定阶。本文对提出的方法进行了仿真研究，结果验证了方法的有效性。

System order determination is an important problem for nonlinear system identification using feedforward neural networks. Nonlinear system model order determinatin using feedforward neural networks is closely related to the generalization of the neural networks. Although OLS algrorithm has been proposed as an good process to construct radial baiss function neural networks, the network constructed by OLS often has poor generalization . ROLS algorithm is, therefore, proposed to combine OLS algorithm with regularization to enhance network generalization. An approach based on ROLS learning algorithm for radial basis function networks is proposed for nonlinear system model order determination. This approach is demonstrated in simulation .