Identificação multiobjetivo de sistemas não-lineares

• Ano de defesa: 2002
• Autor(a) principal: Erivelton Geraldo Nepomuceno
• 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; Engenharia elétrica
• Link de acesso: http://hdl.handle.net/1843/RHCT-5F2HJS

The system identification studies how to model and analyze the data from. In practical situations it is common to collect a limited set of data corrupted by noise and local character. In these situations, if one considers only the data collected, hardly a suitable non-linear model will be obtained. The first attempts to solve this problem using a mono-objective approach, in which the prediction error is the objective to be minimized and auxiliary information is incorporated in the form of constraint optimization techniques using mono-goal. This approach, however, puts into perspective the determination of a set of solutions within which there is a compromise between the various goals. All of these solutions is called Pareto-optimal. This paper employs multiobjective techniques in the identification of nonlinear systems and presents a systematic framework for incorporating auxiliary information. This procedure was called multiobjective identification. The representation NARMAX (AutoRegressive Nonlinear Moving Average model with exogenous input) was chosen to allow the incorporation of ancillary information, particularly concerning fixed points of curve and static. The proposed methodology was applied in the modeling of two chaotic systems (Chua's circuit and the sine map), an electric heater, a buck and a reactive compensator (TCSC the - Controlled Series Capacitor Tryristor or thyristor controlled series capacitor). The final solutions of several problems which belong to the Pareto-optimal in each case were chosen through analysis of preference a priori and a posteriori and two makers: based on a balance bias and variance and another developed in this work, based on minimum standard of objective standard. It was found that the identification of multiobjective nonlinear systems is achievable. The determination of solutions belonging to the Pareto-optimal allows the study, this set, ranging from the way the relevant properties of the models and, based on this analysis, the choice of each model better suited to the specific needs of each situation. The possibility of this analysis is the main advantage of multiobjective methodology