Identifiability of time varying parameters in a Grey-Box Neural Model: Application to a biotechnological process

Grey Box Neural Models (GBNM) constitute a real alternative for those processes for which the available a priori knowledge is incomplete. In this work an application to a biotechnological process has been performed. Good results of the GBNM acting as a software sensor for the non measured state variables has been shown. However even if the estimation performance is good, correct identification of the time varying parameters is not assured. Identifiability of these parameters has to be tested and some proposed techniques are used in this work showing that the specific growth kinetics and the specific production kinetics can be identified although the last one is more difficult because of its dependence on only one measured variable.