Identification of Multivariable NARX Systems Based on Conjugate Gradient Pursuit Algorithm

This paper focuses on the joint parameter and structure identification of the polynomial class multi-input single-output nonlinear autoregressive with exogenous inputs (MISO-NARX) models. A sparse identification method based on conjugate gradient pursuit is proposed by integrating the conjugate gradient optimization algorithm with a greedy strategy. First, a sparse identification model is established according to the order bounds. Then the greedy search along with the conjugate gradient optimization is employed to select and estimate the positions and values of non-zero parameters. The system orders and time lags are derived from the sparse parameter structure. Simulation examples demonstrate that this method effectively leverages limited sampling data to simultaneously identify both the structure and parameters of the system.