Determinants of energy use in UK higher education buildings using statistical and artificial neural network methods

Studies were carried out using Display Energy Certificate (DEC) data for university buildings to investigate building parameters as determinants of energy use. A preliminary statistical analysis was conducted of annual electricity and heating fuel consumption across a UK-wide dataset focusing on building activity and internal environment. Using data for London university buildings only, a pilot study was also undertaken to assess the use of an artificial neural network (ANN) method for analysing a wider range of energy use determinants.For University Occupier Buildings (UOB) it was found that generally electricity use is high and heating fuel use is low relative to the CIBSE TM46 benchmarks for the University campus category. For the London university dataset there was appreciable variation in energy use between different university-specific building activities. Activity was also shown to have a high ANN causal strength together with material, environment and glazing type.Prediction performance of the ANN improved with the addition of building parameters: for electricity use the ANN mean absolute percentage error reduced to 34%, a 30% reduction relative to a theoretical benchmark-based approach; for heating fuel use it reduced to 25%, a 49% reduction against the benchmark-based approach. Prediction performance appeared to be restricted however, perhaps owing to the limited number of training patterns.From the pilot study the ANN methodology appears to be viable for use in analysing building energy use determinants. A broader follow-up study is planned accordingly and various measures to develop the ANN methodology are presented.

► Generally for UK universities electricity use exceeds and heating use falls below TM46 benchmarks.
► For UK universities electricity use typically increases for more intensively serviced buildings.
► Energy use of University Occupier Buildings varies appreciably by building activity.
► Artificial neural networks appear to be viable for assessing building energy use determinants.