Uso de conhecimento prévio na Identificação de modelos polinomiais NARMAX

• Ano de defesa: 1996
• Autor(a) principal: Cristiano Ribeiro Ferreira Jacome
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Engenharia elétrica
• Link de acesso: http://hdl.handle.net/1843/BUOS-8D7ML2

The main objective of this work is to investigate the use of techniques to identify nonlinear dynamical systems with prior knowledge about the system, using NARMAX polynomial models. Such models are parametric input-output structures able to represent dynamic behavior of a wide class of nonlinear systems. The identification of nonlinear parametric models can be divided in five stages. These five stages are introduced and analysed in the scope of nonlinear black-box model identification.Model structure selection is the crucial stage of nonlinear model identification and is the main focus of this work. Thus this issue has been analysed more deeply. The concepts of term clusters and clusters coeficients are used to derive a preliminar procedure to help in the structure selection in nonlinear polynomial models. This procedure associated with prior knowledge about the system being modelled can work efficiently in situations where other techniques fail. In this way, discretized models of linear dynamic systems up to third order with constant gain and invariant time constants are analysed. After that, variations with process signals are included in the gain and one time constant These results are compared with certain NARMAX polynomial models. This procedure is applied to two simulated systems: a Wiener model and a Hammerstein model. Finally, the developed procedure is applied in the identification of a real control valve. The valves static nonlinearity is recovered from the identified dynamic model. Besides this practical application, it will be seen that is possible to get reduceddynamicaly valid NARMAX models, under certain assumptions and prior knowledge about the system. The importance the new results become evident in situation when usual identification techniques can not improve the model structure any further.