Anomaly Detection in Energy Performance Certificates – From Oblivious to Enlightened
Résumé
Energy performance of buildings is central to achiev- ing the climate ambition, and is part of the directives of many countries. Energy Performance Certificates (EPCs) are an integral part of such directives, and are an important instrument to monitor and improve the energy performance of the buildings. However, some EPCs may suffer from inconsistent information and anomalies, making them inappropriate for use. Anomaly detection is a long standing and crucial problem in dependability. The challenge of detecting anomalies in EPCs is twofold: on the one hand, the causes of anomalies are very diverse and not all known a priori, and on the other hand, there is no EPC data labeled with information about their (ab)normality. In this paper, we first share our experience in applying several oblivious anomaly detection methods to EPCs. We then devise an iterative cross-methods approach for a more efficient EPC anomaly detection. Experiments on four real-world EPC datasets show the effectiveness of the proposed anomaly detection approach, and how it outperforms state-of-the-art techniques.