Título: Identificação de Processo Não-linear Usando Rede Neural com Método de Treinamento Híbrido Baseado em Filtro de Kalman e Máxima Descida
Autores: Coelho, Leandro dos Santos; Villa, Luiz Fernando
Resumo: There are a wide variety of identification methods and model structures forms to choose for system nonlinear identification applications. Neural networks have proved to be versatile and useful nonlinear models for industrial processes. This paper presents a comparative study of learning methods for training of radial function neural networks (RBF-NN). Simulations involving the RBF-NNs with training methods based on Kalman filter, gradient descent, and a Kalman-gradient descent hybrid approach. The simulation results indicate the potentialities of the hybrid approach in RBF-NNs learning for one step ahead identification of a pH neutralization case study.
Código DOI: 10.21528/CBRN2005-013
Artigo em PDF: CBRN2005_013.pdf
Arquivo BibTex: CBRN2005_013.bib