Dynamic Systems Numerical Integrators in Neural Control Schemes

Título: Dynamic Systems Numerical Integrators in Neural Control Schemes

Autores: Rios Neto, Atair

Resumo: In this short paper, a not yet explored way of getting dynamic systems discrete forward models, to be used as internal models in control schemes, is proposed. A mixed heuristic and theoretical approach is taken to propose and explicitly show how to use dynamic systems ordinary differential equations (ODE) numerical integrators in control schemes where a discrete forward internal model is needed. Use is made of the structure of numerical integrators algorithms to make it possible to get a neural feedforward model to approximate the dynamic system by only learning the derivative function in the system ordinary differential equations model. It is then illustrated how to use this kind of discrete forward model in a predictive control scheme and in an internal model control scheme where a least control action criterion is used. Independently of any numerical experiment, conclusions are drawn concerning the peculiar and advantageous aspects of the proposed method.


Páginas: 4

Código DOI: 10.21528/CBRN2001-019

Artigo em pdf: 5cbrn_019.pdf

Arquivo BibTex: 5cbrn_019.bib