METODOLOGIA NEBULOSA PARA IDENTIFICAÇÃO RECURSIVA NO ESPAÇO DE ESTADOS BASEADO EM AGRUPAMENTO EVOLUTIVO DE DADOS.

Abstract

In this dissertation, an evolving fuzzy methodology for the identification of nonlinear systems is proposed. The obtained evolving model is capable of automatically adjust its structure according to the data flow. In addition, the minimum realization of consequent part of the fuzzy rule ensures the simplicity of the obtained model. In order to compare the proposed methodology with other existing techniques in the literature, the identification of two benchmarks used in other works widely cited in the literature was carried out. The results obtained were competitive and advantageous in relation to the methodologies used in the comparison. The proposed technique was successfully applied in the modeling of a 2DoF Helicopter. This system represents a complex challenge for identification methodologies because of its high level of complexity. The results obtained demonstrated the ability of the proposed methodology to represent real systems of high complexity. To demonstrate the applicability of the evolving fuzzy methodology for recursive state space identification, it was proposed the estimation of the trajectory of a rocket used for training. The results obtained were encouraging and demonstrated the applicability of the proposed methodology in applications with a high level of complexity. Due to the evolving nature of the proposed methodology, a good estimation of the rocket’s trajectory during its flight time was possible. This achievement is due to the ability of the evolving model to adapt to the data set in an online way, thus guaranteeing good results during all stages of flight.