Correntropy‐Based Robust Interval‐Varying Recursive Estimation Method for Nonlinear Systems With Spline Networks and Outliers

ABSTRACT Nonlinear system identification plays a key role in real‐world modeling. The spline networks can model the nonlinearity with a high precision without prior knowledge of the nonlinear structure. This paper examines the identification problem of the Hammerstein nonlinear systems with outliers by using spline networks to describe the nonlinearity. To avoid the redundant computation, two identification sub‐models are derived, one with local nonlinear spline parameters and the other with global linear parameters. By exploiting the insensitivity of the correntropy to outliers, a correntropy‐based robust interval‐varying recursive estimation method is presented. The proposed method not only models the unknown nonlinearity with a high computational efficiency but also is robust to outliers under the premise that the total distribution of the observed data is unknown. The superiority of the proposed algorithm is validated by simulation experiments.