Identification of non-linear dynamic systems in power plants

Problems related to the identification of non-linear systems are analysed by considering as a case study the neural modelling of the furnace and the superheater systems. As far as the furnace is concerned, identification addresses neural modelling of the total heat reaching the evaporator; the process is not dynamic because the heat generation is particularly rapid. Conversely, this is not the case in a superheater where dynamics play a relevant role: identification of the steam and the flue gas temperatures requires specific recurrent type neural models. Identification of such systems, belonging to a one-through 320 MW group, are the first step in developing computationally simple distributed nonlinear neural models for the whole plant. Issues related to training data extraction, training algorithms and stability are taken into account.