Identification of NARX Model Based on Model Predicted Error for Cantilever Beam

In this paper, a novel identification method for Nonlinear Autoregressive with eXogenous Inputs (NARX) models based on the model predicted output (MPO) error is proposed. Due to the limited spectral complexity of harmonic excitation signals, the NARX model obtained from the identification under harmonic excitation often fails to achieve the desired accuracy and sometimes results in nonconvergence. In this paper, a NARX model identification algorithm is proposed. This algorithm optimizes the potential identification results based on the MPO error, and the modeling effectiveness of the proposed algorithm and the traditional identification algorithm is compared through simulation and experimentation on cracked cantilever beams. Simulations and experiments demonstrate that the proposed method reliably improves modeling accuracy even under harmonic excitation with limited spectral complexity.