Using multiple linear regression to disaggregate electricity consumption for cluster-metered academic buildings

Ryerson University does not have a means to gauge electricity consumption for half of their campus buildings. The installation of utility meters is outside of the University’s budget, a situation that may be similar across other academic institutions. A multiple linear regression approach to estimating consumption for academic buildings is an ideal tool that balances performance and utility. Using 80 buildings from Ryerson University and the University of Toronto, significant building characteristics were identified (from a selection of 18 variables) that show a strong linear relationship with electricity consumption. Four equations were created to represent the diversity in size of academic buildings. Tested using cross-validation, the coefficient of variation of the RMSE for all models was 33%, with a range of error between 20% and 43%. The models were highly successful at modeling electricity consumption at Ryerson University with an average error of 14.8% for five building clusters. Using metered data from each cluster, raw estimates for individual buildings were adjusted to improve accuracy.