Combining first principles with grey-box approaches for modelling a water gas heater system

This paper presents two different approaches for parameter identification in Hybrid Models (HM). The hybrid model consists in the parallel or cascade connection of two blocks: an approximate First Principles Model (FPM) and an unknown block model. The first principles model is constructed based in the balance equations of the system that could have some unknown parts. The unknown parts of the global model are modelled with two grey-box model techniques: neuro-fuzzy systems and nonlinear auto regressive models. These two types of hybrid models generate models with high interpretability degrees. The Nonlinear Auto Regressive Hybrid Model (NARXHM) proposed uses a NARX model, tuned with the least squares algorithm, in cascade, with a first principles model. The Neuro-Fuzzy Hybrid Model (NFHM) proposed uses a neuro-fuzzy model, tuned with the gradient decent combined with least square algorithms, in cascade with a first principles model. The identification of the two hybrid model structures are implemented off-line. Finally, to test and compare these two transparent modelling strategies, they are applied for modelling a water gas heater non-linear system.