Energy benchmarking are used to compare the operational performance of buildings with the corresponding stock. Multi-criteria methods emerged to consider different factors in benchmarking assessment. However, there is a lack in considering occupants’ thermal satisfaction in methods based on actual data. The objective of this article is to propose a method to integrate thermal satisfaction into energy benchmarking. The main innovation is to propose a probabilistic metric that takes into account energy consumption, construction aspects, climate conditions, systems and thermal satisfaction level to benchmark a building. The method consists of a statistical analysis to select relevant variables in the building stock, the process of discretisation of such variables, and the developing and validation of a Bayesian Network to serve as an instrument for the benchmarking method. A detailed evidence-based dataset of 426 schools in Brazil was used. Results showed that buildings with low thermal satisfaction of occupants were benchmarked as less efficient than those with high thermal satisfaction and similar energy consumption. Regarding the validation step, the benchmarking model achieved an error rate ranging from 17.78% to 29.17%. The main conclusion is that machine learning techniques can adequately integrate subjective aspects such as occupant satisfaction in data-driven energy benchmarking methods.
Geraldi, M.S.; Ghisi, E.