Identificação de modelos de Hammerstein e Wiener para sistemas não lineares multivariáveis utilizando métodos de subespaços

Abstract

The subspace methods for system identification are a powerful tool for modelingvmultivariable linear systems in state space. These methods do not require that all statesvare directly measured. Combining the methods of subspace to other techniques, it isvpossible to identify nonlinear dynamical systems by means of multivariable models ofvinterconnected blocks. These models consist of a linear dynamic block interconectedwith static nonlinear curves. However, the literature on this form of modeling appliedvto multiple input and multiple output systems (MIMO) is still scarce. Thus, this workvaddresses this topic and has the objectives of (i) investigating the existing subspacevmethods for both SISO and MIMO nonlinear systems and of (ii) proposing methods forvthe identification of nonlinear MIMO systems, using block-structured Hammersteinvand Wiener models. The nonlinear static block of the interconnected block modelsvis determined by two dierent methods: static tests and harmonics tests. The linearvdynamic block of models is estimated by the classical MOESP subspace method. The investigated methods are tested in simulated examples in order to illustrate the particularities of each technique as well the extension of their applicability.