Monitoring abnormal energy consumption is helpful for demand-side management. This paper proposes a framework for contextual anomaly detection (CAD) for residential energy consumption. This framework uses a sliding window approach and prediction-based detection method, along with the use of a concept drift method to identify the unusual energy consumption in different contextual environments. The anomalies are determined by a statistical method with a given threshold value. The paper evaluates the framework comprehensively using a real-world data set, compares with other methods and demonstrates the effectiveness and superiority.
|Conference||International Conference on Neural Information Processing (ICONIP 2020)|
|Period||18/11/2020 → 22/11/2020|
|Series||Neural Information Processing - Letters and Reviews|