Campus Building Energy Usage Analysis and Prediction: A SVR Approach Based on Multi-scale RBF Kernels

Building energy usage analysis and prediction has been very important to energy conservation and occupant comfort optimization. This paper focuses on hourly air conditioning energy usage for teaching buildings on campus. First, a detailed analysis is made on real world energy usage data from campus energy monitoring platform. Then an adaptive 24-hour ahead prediction model is proposed based on Support Vector Regression (SVR). We investigate the feasibility of a new kernel obtained by a linearly weighted combination of multiple radial basis functions (RBF). A problem-based particle swarm optimizer (PSO) is further implemented to facilitate model parameter estimation. The results of our experimental study indicate that the proposed model is a promising alternative in modeling hourly air conditioning energy performance for campus buildings.