Identificação de Sistemas Não Lineares a partir de Modelos de Hammerstein Utilizando Aprendizagem Profunda

Abstract

In this work, connections between the areas of systems identification and deep learning are explored. A method of nonlinear systems identification is proposed based on deep neural networks and Hammerstein models. The proposed approach is formulated from the separation of the linear and non-linear parts of the models. Thus, the used network structures reflect this separation strategy and the identification is performed through the training of deep neural networks. As a reference, was adopted an identification method in which orthonormal and radial basis functions are used. In general, the proposed approach was able to provide better identification results compared to the adopted reference method.