Data-Driven Fuzzy Modelling Methodologies for Multivariable Nonlinear Systems

Abstract

In this paper, two methodologies of data-driven fuzzy modelling for multivariable nonlinear systems based on Observer/Kalman Filter Identification (OKID) and the Eigensystem Realization Algorithm (ERA) are proposed. The multivariable nonlinear system is represented by a fuzzy Takagi-Sugeno (TS) model, whose antecedent is constituted by linguistic variables (fuzzy sets) and the consequent is constituted by linear submodels in state-space discrete representation. The antecedent parameters are obtained using clustering fuzzy algorithms and the consequent parameters (state matrix, input matrix, output matrix and direct transition matrix) are obtained using the algorithm discussed in this article. Experimental results for identification of a Quadrotor Unmanned Aerial Vehicle (UAV) are presented, in order to illustrate the efficiency and applicability of the methodologies in real systems with coupled data and real systems with decoupled data.