Multivariate stochastic modelling of a full-scale office zone and air conditioning plant: The potential for BEMS

Abstract

In facilities management, mathematical models have the potential to improve system management, to enhance control, and to maintain healthy conditions for occupants of buildings. Various model types and implementation methods exist, and to date these have chiefly been investigated using simulation and/or laboratory small-scale set-ups. In this paper, the application of the identification technique to full-scale system modelling is demonstrated. Multivariable stochastic models are identified which describe the thermal and moisture behaviour of air in a full-scale unfurnished office zone serviced by a dedicated air-conditioning plant. The models are used to forecast values af air temperature and relative humidity, and are validated by comparing their short-term (10-20 min) and long-term (several days) forecasts with measured data. Short-term prediction accuracies were found to be well within ±0.25°C in 19°C and ±0.6% RH in 53% RH, while corresponding long-term values were to within ±0.8 °C and ± 1.3% RH, respectively. The results show that good quality models are obtainable for full-scale systems, while comparisons with a simple piece-wise constant approximation (next value = current value) further demonstrated the superiority of the model-based approach. The implications for the future development of bems are also discussed.