PSF+ — Fast and improved electricity consumption prediction in campus environments

Smart meters are becoming ubiquitous around the world, and their use is even being legally mandated in some countries. This increased use of smart meters has led to a flurry of research activity surrounding the analysis of smart meter data. However, much of this work relates to consumer smart grids, and there exists limited work on analysing data in a campus-style environment. In this work, we illustrate the unique challenges next-day electricity load consumption forecasting in a heterogeneous campus environment entails, and develop methods to enhance a cluster-based forecasting technique to address these challenges. We develop a method to easily incorporate external information to the predictor, and also develop an ensemble of predictors which significantly improves prediction accuracy. We evaluate our methods using three real-world datasets obtained from smart meters of the campuses of the University of Melbourne, University College London and Cornell University. Our results show significant improvement in accuracy over all three real-world datasets, both in comparison to the original technique as well as other competitive linear and non-linear methods.