Data Driven System Identification Using Evolutionary Algorithms

Abstract

We present an evolutionary algorithm(EA) based system identification technique from measurement data. The nonlinear optimization task of estimating the premise parameters of a Takagi-Sugeno-Kang fuzzy system is achieved by a EA, the consequent parameters are estimated by least squares. This reduces the search space dimension leading to greatly reduced load on the EA. The significant contribution of this work is in formulating the fitness function that judiciously applies selection pressure by 1) penalizing low firing strengths of rules, and, 2) by penalizing low rank design matrix at the rule consequents. The proposed method is tested on the identification of non-linear systems.