含有色噪声的神经模糊Hammerstein模型分离辨识

Separation identification of neuro-fuzzy Hammerstein model with colored noise

针对实际工业过程中普遍存在的有色噪声,本文提出一种基于递推增广最小二乘算法的神经模糊Hammerstein模型辨识方法,突破了传统的Hammerstein模型迭代分离算法.首先,利用多信号源实现Hammerstein模型中静态非线性环节和动态线性环节的分离,大大简化了辨识过程,提高了串联环节参数的分离精度.其次,利用长除法将噪声模型用有限脉冲响应模型逼近,采用增广递推最小二乘法进行线性环节的参数估计.最后,采用神经模糊模型拟合静态非线性环节,同时设计了神经模糊模型参数的非迭代优化算法,改善了模型的使用范围.该方法保证了模型的预测精度,对含有色噪声的非线性系统具有较好的拟合效果.仿真结果验证了上述方法的有效性.

To deal with the colored noises commonly existing in practical industrial processes, we propose an identi- fication method for neuro-fuzzy Hammerstein model using the recursive extended least squares algorithm （RELS）. This method is different from the traditional iterative separation methods for identifying Hammerstein model. Firstly, multiple signal sources are employed to separate the static nonlinear part and the dynamic linear part of the Hammerstein model in order to simplify the identification process and improve the accuracy of the model parameters. Secondly, the finite impulse response model is used to approximate the colored noise model by using the long division method, Then, parameters of the linear part are estimated by using the RELS algorithm. Finally, the neuro-fuzzy model is adopted in identifying the static nonlinear part. Meanwhile, a noniterative neuro-fuzzy optimization algorithm which can be applied to many nonlinear systems is designed. The proposed method can guarantee high precision of the Hammerstein model. Moreover, it has the ability to approximate the nonlinear systems with colored noises. Simulation results show the effectiveness of the proposed method.