Input Design and Recommendations for the Identification of Hysteretic NARX Models
Keywords: Design Excitation Inputs, System Identification, NARX models, Hysteresis
AbstractThis paper deals with important aspects in the identification of Nonlinear AutoRegressive models with eXogenous inputs (NARX), with a particular attention for hysteresis. The main contribution is a mathematical formulation for the design of excitation inputs that can preserve the frequencies of interest and force the system to achieve different operating points. Other important contribution is the discussion about the influence of some identification meta-parameters and structural constraints under the predictive performance of NARX models in the hysteretic context. The identification procedure is illustrated with two numerical examples and one experimental, a pneumatic valve with hysteresis-type nonlinearity. For each example, a NARX model is identified using classical methods. Due to their simplicity, these models are promising for applications and they will be used for nonlinearity compensation in future works.