Identificação de sistemas usando NARX com termos

Abstract

Linear models have been widely described in the literature and have consolidated analysis and control methods. These models are characterized by their simplicity, efficiency, sta- bility and robustness, but this class is not capable of faithfully representing more complex systems. In addition, the development of new mathematical representations, advances in computing power and the increased demand for more ccurate estimates have motivated the study of non-linear models. This work addresses the offline identification of systems using a gray box approach. Where the polynomial NARX model was altered so that new regressors generated from the inputs and outputs were added. Its representation remained linear in parameters, so the least squares method was used for estimation. Meanwhile, structure selection was carried out using the SEMP and FRP algorithms. The case studies covered measurements on a system of connected tanks, a benchmark of motors coupled to the pulley and a numerical example. Finally, the models obtained were simulated, validated and compared with their polynomial counterpart, in order to verify the gain from applying non-linear functions to create new regressors.