Transformer-Based Anomaly Detection in Energy Consumption Data

Dimitrios Karpontinis, Georgios Alexandridis

Abstract: As electric energy demand continues to rise, understanding electricity consumption habits and promptly detecting abnormal usage patterns are crucial steps in reducing energy waste. This study introduces a novel anomaly detection approach that identifies out-of-the-ordinary patterns in energy consumption, leveraging state-of-the-art deep learning techniques on relevant data and more specifically, a transformer-based architecture. By utilizing the proposed methodology, the energy demand is classified either as expected or unexpected, based on the discovered energy patterns. The latter cases can then be further examined in order to find the potential causes for such anomalies. In this respect, the proposed architecture can be used to gain insight into electricity consumption patterns and actively work towards reducing energy waste.