6 - Model identification

Abstract

This chapter describes an overall framework for model identification. This process enables us to achieve a complete model, that is to say that we have both a model structure and a fully determined set of parameters corresponding to that structure. By this stage of the modeling process at least one candidate model is achieved. It is equally possible, however, that more than one could also be achieved, with the need to choose between them as to which is the most appropriate. Focusing on a single model, if it is incomplete this will be due to some of the parameter values being unknown. This is true when the modeling approach has been driven either by data or by the physiology of the system. So one may deal with the whole model or just part of it. This chapter aims to provide a framework for dealing with this situation. To solve this identification problem, as it is known, we need data. Data sometimes occur from the intrinsic dynamics of the system, for instance, in the form of spontaneous oscillations or noise. The selection of appropriate test signals is of essential importance to the identification process. The most commonly applied test signals in metabolic and endocrine studies and in studies on physiological organ systems are those that result in a transient response of one or more system variable. In general, test signals can be classified into those that result in a transient response, those that result in harmonic signal analysis (frequency response), and those that involve random signal analysis