A new strategy to benchmark and evaluate building electricity usage using multiple data mining technologies

This study presents a new strategy using cluster analysis, multivariate adaptive regression splines and conditional inference trees to benchmark and evaluate building electricity usage. Different from the existing studies, cluster analysis was first used to group buildings based on their annual electricity usage patterns in order to improve the accuracy of the benchmarking result. Multivariate adaptive regression splines technique was then applied to capture the non-linear relationships (i.e. benchmarking models) between building electricity usage per square meter and explanatory factors with enhanced interpretability. Conditional inference trees technique was further used to evaluate the benchmarking result. The performance of this strategy was evaluated using two-year time-series electricity usage data of 20 university buildings. The results showed that the multivariate adaptive regression splines can effectively describe the complex non-linear relationships between building electricity usage per square meter and explanatory variables. The conditional inference trees can help identify conditions when buildings had higher electricity usage. The results obtained from this study can be further used to assist in building energy auditing, and fault detection and diagnosis.