Título: Uma Rede Neural Construtiva Com Atualização Dinâmica De Pesos
Autores: Valença, Mêuser; Ludermir, Teresa
Resumo: This paper is concerned with the applications of the GMDH to prediction models of river flow and exchange rates. Both applications are associated with the modeling of environmental, ecological, and social systems which have a great deal of uncertainty and unknown iterations between elemets and parameters. An adaptive scheme of the prediction model is constructed by making use of the modified algorithm GMDH (multiplicative-additive algorithm) to improve the forecasting ability when a small number of data sequences are available for the model building. The GMDH algorithms have, of course, both strengths and weaknesses compared with other prediction models. Our self-organizing method, GMDH, has the specific features of algorithmic, short computation time, and applicability to a limited amount of data.
Artigo em pdf: 4cbrn_027.pdf
Arquivo BibTex: 4cbrn_027.bib