Data-driven modelling of a R744 refrigeration system with parallel compression configuration

Abstract

The type and characteristics of the employed components, subsystems and refrigerant in a commercial refrigeration system varies significantly depending on required functionalities (e.g. MT (medium temperature), LT (low temperature), AC (air conditioning)) and external factors (e.g. ambient temperature). Thus, modeling for control-oriented objectives becomes a challenging task. As these systems provide a large amount of data, modelling for control-oriented objectives based on a data-driven method become an appealing option. This work focuses on obtaining a (controloriented) data-driven MIMO (multiple input multiple output) model of a CO2 booster refrigeration system, supported by parallel compression and ejectors, producing refrigeration for MT, LT, AC. Subspace identification is used for obtaining the data-driven MIMO model. The data-driven modelling approach is validated with data from a CO2 system where negligible deviations are observed even with change in the main disturbances.