ADEPT: Anomaly Detection, Explanation and Processing for Time Series with a Focus on Energy Consumption Data

Anomaly detection techniques are applicable for recognizing excessive energy consumption and device failure, thereby contributing to the maintenance of operational and sustainable energy supply systems. In this context, human decision makers can benefit from receiving explanation attempts for detected anomalies as part of a semi-automated software solution. Therefore we introduce the framework ADEPT, which comprises interfaces for processing user-supplied time series data and interactively visualizing explanatory anomaly information. Our framework features several shallow and deep machine learning algorithms for anomaly detection and explanation. We demonstrate ADEPT using energy consumption data collected from our university campus.