Aplicação de Máquinas de Aprendizado Extremo ao Problema de Identificação de Sistemas Não-Lineares

This work proposes the application of extreme learning machines to the problem of nonlinear system identifi- cation. Such structure constitutes a promissing option because it presents the capability of approximating nonlinear mappings and simplicity in the process of parameter adaptation. The proposed strategy is analyzed in a scenario related to the control system of a pneumatic valve, being compared with two other models based on multilayer perceptron networks. The obtained results show that the ELM is very suitable for the task.