Thermal behaviour model identification for an office space using BMS data

In this study, linear Box—Jenkins, output-error and non-linear neural network autoregressive NARX models are investigated to predict the thermal behaviour of an office positioned in a modern commercial building. External and internal climate data recorded over a summer season were used to build and validate models. The paper exploits the potential of using linear and non-linear models to predict room temperature at different time scale ahead (5 min or 4 h ahead). The prediction performance is evaluated using the criteria of goodness of fit, errors and mean-squared error between predicted model output and real measurements. The results demonstrate that all models provide reasonably good predictions but non-linear models outperform linear models.

Practical application: Prediction of room temperature by black-box linear and non-linear models obtained can be utilised in the building temperature control strategy. When there is any change in the building thermal behaviour (e.g. more equipments added or office equipment re-arranged), the performance of traditional non-adaptive building temperature proportional—integral (PI) or proportional—integral—derivative (PID) controls will deteriorate. The models based on immediate past records and actual behaviour can be adapted to any changed. The PI and PID controller with such models integrated are adaptable to any changes and hence maintain performance. The modelling techniques studied in here are not restricted to office building temperature prediction and control problems; they can be extended to other types of buildings such as hospitals, supermarkets, airports and schools.