Controle de processos com redes neurais inversas

Abstract

The use of artificial neural networks in control process had its great pulse starting from the decade of 80, when several use proposals appeared. This work seeks to do a comparison among that technique, considered recent, with the conventional denominated techniques of the decade of 40. For so much, we accomplished the comparison among the a controller's acting based on nonlinear artificial neural network and a controller PID. We also compared two training types involving the nonlineal artificial neural network, Steepest Descent and Decomposition in Singular Values. We described in this work the steps used for those comparisons, beginning with the development of the control software, implementation of the used controllers' algorithms and assembly of a module of tests. They are also presented details of the control software, of the training types and of the characterization of the test module. The advantages, disadvantages and difficulties of implementation, so much in the simulate way as in the module of tests, are described in this work always highlighting which the controller that obtained better performance.