Introduction to Block-oriented Nonlinear Systems

Abstract

System identification refers to the experimental approach that consists of determining system models by fitting experimental data to a suitable model structure [14] in some optimal ways. Linear model structures can be based upon when the physical system remains in the vicinity of a nominal operation point so that the linearity assumption is satisfied. When a wide range of operation modes are involved, the linear assumption may not be valid and a nonlinear model structure becomes necessary to capture the system (nonlinear) behaviour. In relatively simple cases, suitable nonlinear model structures are obtained using the mathematical modelling approach that consists of describing the system phenomena using basic laws of physics, chemistry, etc. Then, system identification methods may be resorted to assign suitable numerical values to the (unknown) model parameters. When the mathematical modelling approach is insufficient, system identification must rely on ‘universal’ black-box or grey-box nonlinear model structures. These include NARMAX models [9], multi-model representations [15], neuro-fuzzy models [3], Volterra series [19], non-parametric models [14] and others.