Evolutionary Neuro-Computing Approaches to System Identification

Abstract

System models are essentially required for analysis, controller design and future prediction. System identification is concerned with developing models of physical system. Although linear system identification got enriched with several useful classical methods, nonlinear system identification always remained active area of research due to the reason that most of the real world systems are nonlinear in nature and moreover, having non-unique models. Among the several conventional system identification techniques, the Volterra series, Hammerstein-Wiener and polynomial model identification involve considerable computational complexities. The other techniques based on regression models such as nonlinear autoregressive exogenous (NARX) and nonlinear autoregressive moving average exogenous (NARMAX), also suffer from dfficulty in choosing regressors.