Demand Forecasting Models for Academic Campus Buildings

The use of demand forecasting, benchmarking and response software is proposed as a solution for the Facilities Management division at Stellenbosch University to reduce the maximum demand cost component of campus facilities. This paper aims to address the demand forecasting segment by investigating three commonly used forecasting methods, namely Least squares estimation (LSE), Gaussian process regression, and Artificial neural networks. Forecast models implementing these methods are designed, trained on historical demand data and weather data from 2013 to 2015, and their response for 2016 is analyzed. The LSE forecast model yields the most accurate results for the following input parameters: forecasted temperature and ultraviolet A-rays (UVA), time of the day, and the corresponding demand measurements of one and two weeks earlier.