Identification of combined physical and empirical models using nonlinear A priori knowledge

Abstract The incorporation of a priori knowledge in process identification is discussed. Using a form of prefiltering, referred to here as data extraction, the effects of dynamics known a priori are removed from available input/output process data. The resulting information is then used to identify the remaining unknown process dynamics. This approach to system identification has important advantages — particularly with regard to noise sensitivity, necessary amount of excitation, and the complexity of the model to be estimated. The data extraction method in its nonlinear version is applied in simulations and to actual data from a bench-scale and an industrial process.