A neuro-fuzzy identification of non-linear transient systems: Application to a pilot refrigeration plant

Abstract This study proposes to determine empirical models that represent the non-linear dynamics of a pilot refrigeration system (chiller), by using identification techniques of neuro-fuzzy systems (ANFIS). A detailed experimental study of a pilot refrigeration process using compression was conducted. Non-linearities were detected in the experimental data obtained from the system. Neuro-fuzzy models were developed for the prediction of the following process temperatures: condensation; evaporation; and refrigerated fluid (propylene glycol aqueous solution). The validation of the models was successfully conducted offline and online. The values obtained for VAF (“variance accounted for”) performance index and the dispersion plots showed the high performance of the models. These empirical models will be especially useful in the development of different non-linear control strategies, such as: fuzzy, neuro-fuzzy and model predictive control.