Attention Boosted Autoencoder for Building Energy Anomaly Detection
This work addresses energy conservation in buildings by enabling early detection of operational deviations, though it appears incremental as it builds on existing autoencoder-based anomaly detection methods.
The paper tackles building energy anomaly detection by proposing a novel attention mechanism to model consumption behavior, demonstrating its effectiveness on a real-world dataset with sample case studies and visualization.
Leveraging data collected from smart meters in buildings can aid in developing policies towards energy conservation. Significant energy savings could be realised if deviations in the building operating conditions are detected early, and appropriate measures are taken. Towards this end, machine learning techniques can be used to automate the discovery of these abnormal patterns in the collected data. Current methods in anomaly detection rely on an underlying model to capture the usual or acceptable operating behaviour. In this paper, we propose a novel attention mechanism to model the consumption behaviour of a building and demonstrate the effectiveness of the model in capturing the relations using sample case studies. A real-world dataset is modelled using the proposed architecture, and the results are presented. A visualisation approach towards understanding the relations captured by the model is also presented.