Metodologia de identificação nebulosa evolutiva multivariável no espaço de estados com ordem variante no tempo

Abstract

In this work, an evolving fuzzy methodology for the identification of nonlinear, Multi-Input Multi-Output (MIMO) dynamic systems, based on the algorithms OKID (Observer/Kalman Filter Identification) and ERA (Eigensystem Realization Algorithm), is proposed. The evolving model obtained is capable of autonomously changing its structure according to the data flow. The OKID algorithm is executed recursively, not requiring the storage and batch processing of input and output data of the real system, significantly reducing the need for memory allocation and execution time. The minimal realization of the rule consequent submodels guarantees the simplicity of the model obtained. Moreover, the local state space models are able to change their order independently for each fuzzy rule. In addition, this methodology aims to reduce the number of algorithm parameters to be specified during its initialization. The proposed technique was applied in the identification of a tank system, an aerial quadrotor robot, an insdustrial evaporator, and a glass kiln. The results obtained demonstrate the applicability of the proposed methodology.