Redes neurais aplicadas em estrategias de controle não linear

AUTOR(ES)
DATA DE PUBLICAÇÃO

2002

RESUMO

The artificial neural networks are computational tools with a great number of applications in modeling techniques and process control. Such fact is due its capacity to learn sufficiently accurate models and give good nonlinear control when model equations are not known or only partial state information is available. Neural network approach allows taking into account in an elegant and adequate way process non-linearities as well as variable interactions. The developed work explores the use of the neural networks in multivariable control strategies as dynamic models for predictions as well as in the definition of the control strategy based on neural networks with on-line learning. The off-line learning of the neural networks is accomplished with a consistent group of historical data of perturbations and responses of the process, which should guarantee at least a satisfactory performance of the neural network for starting of the control system. It is also explored the use of static models of the process, based on neural networks, coupled with a on-line optimization routine, objectifying to identify the best operational conditions to attend specifications of the process. In this context, multivariable control strategies were developed exploring the potentialities of the neural networks as process model and/or as controllers, emphasizing the on-line learning. Several computational programs were implemented for the developrnent of this work in Fortran 90 language program relative to the control algorithms proposed and evaluated. The obtained results show the efficiency of the approached techniques, checking the potential of the use of the neural networks in control strategies

ASSUNTO(S)

controle de processos quimicos redes neurais (computação)

ACESSO AO ARTIGO

Documentos Relacionados