Data-driven modelling of a commercial cold storage system using subspace system identification

• A novel data-driven model of a commercial cold storage system that incorporates external temperature as input is developed using subspace system identification. • A fourteen-day outdoor temperature between 8.9°C and 32.8°C that depicts the temperature of Jos, Plateau State, Nigeria was extracted from a weather profile for Phoenix, Arizona, USA to simulate realistic outdoor temperature for the modified model to generate synthetic data for the estimation and validation of a linear state-space model. • The stability of the identified model is enforced using simulation focus and the model is suitable for testing linear control algorithms. • Results show that the best identified model has a goodness of fit of 98.66% and 90.42% for both outputs, final prediction error of 4.11e-15 and mean square error of 0.0005660. • The data-driven model presents a holistic picture of the whole system that is stable, robust and suitable for testing linear control algorithm. This study presents subspace system identification of a cold storage system incorporating external temperature as input. The proposed model presents a holistic view of the whole system with each subsystem cohesively linked together. A high-fidelity simulation benchmark model of a supermarket refrigeration system from Aalborg University, Denmark was modified by removing open display cases due to their inefficient operation. The modified benchmark model consists of a cold storage room represented as a closed display case, the suction manifold and the compressor rack. A fourteen-day outdoor temperature between 8.9°C and 32.8°C that depicts the temperature of a tropical climate was extracted from a weather profile for Phoenix, Arizona, USA to simulate realistic outdoor temperature for the modified model to generate synthetic data for the estimation and validation of a linear state-space model. The data of the expansion valve, suction pressure, compressor capacity, heat transfer rate and the ambient temperature were taken as inputs while the data of the air and goods temperatures were taken as outputs to achieve a holistic picture of the entire system. Results show that the best identified model has a goodness of fit of 98.66% and 90.42% for both outputs, final prediction error of 4.11e-15 and mean square error of 0.0005660. It also has a model order of 7, thereby giving the best trade-off between accuracy and complexity. The proposed model is stable, robust and suitable for testing linear control algorithms.