Robust Estimation for Hammerstein Models Based on Variational Inference

Abstract

The paper presents a robust identification method using variational inference (VI) for Hammerstein models in the presence of process noise and non-Gaussian colored measurement noise. First of all the measurements and process output are described as Student’s t and Gaussian distribution by using introduced variational parameters. Then the conjugate prior information of introduced parameters is framed for sake of a closed-loop solution. By applying the idea of VI, estimates of system parameters are got by minimizing Kullback-Leibler (KL) divergence. Finally, a numerical simulation example is used to show the effectiveness of the proposed identification method compared with the traditional method.