Applying machine learning to develop energy benchmarking for university buildings in Brazil

University campuses are comprised of many different types of buildings and thus, to improve their energy efficiency, the benchmarking of these buildings is fundamental. In this regard, this study aimed to develop an energy consumption benchmark for university buildings in Brazil. The benchmark was based on a database obtained through parametric building energy simulation by employing an archetype of a university building, developed by the Brazilian Council for Sustainable Construction (CBCS), and a set of parameter variations. The database was obtained considering nine building parameters (e.g., occupancy, internal loads and envelope features), resulting in 23,256 cases aimed at representing the variety of university buildings countrywide. Therefore, this process is innovative since a benchmark is developed for multiple buildings across the country rather than for only one building on a specific university building or campus. Three machine learning techniques were compared to develop the benchmark, multiple linear regression (MLR), support vector machine (SVM) and artificial neural network (ANN). The SVM method had the lowest mean absolute error, root mean absolute error and the highest R2 value, and thus it was adopted to develop the benchmark and efficiency scales. The efficiency scale was used to classify the buildings into 'efficient', 'typical' or 'inefficient', and supports the identification of good practice or inefficiency.