Identificação de sistemas não lineares usando modelos NARX polinomiais

Abstract

his work presents the analysis, implementation, and comparison of two system identification methods based on the polynomial NARX model: FROLS and SEMP, applied to three case studies. The FROLS method employs a forward selection approach, constructing an orthogonal set of regressors and selecting, at each step, the one most correlated with the system output. In turn, the SEMP method simulates possible model combinations and includes only the regressors that significantly contribute to error reduction, discarding redundant ones. The comparison was carried out through the identification of three nonlinear SISO dynamic systems, using RMSE as the performance metric, based on the difference between the expected output and the estimated output in free-run simulation. The results indicate that both methods are efficient and robust, with SEMP showing superior performance in two out of the three cases analyzed.