<![CDATA[Household electricity consumption often exhibits sudden and unexplained spikes that typically go unnoticed until the monthly bill arrives. These anomalies may stem from equipment malfunction, inefficient appliance usage, or irregular electrical patterns that households cannot easily observe. This study proposes an unsupervised anomaly detection framework based on autoencoders to identify abnormal consumption behavior from high resolution household electricity meter data. The model learns normal consumption patterns through reconstruction and flags anomalies using a dynamic threshold derived from reconstruction error distribution. Experimental results demonstrate strong detection capability, particularly for sudden spikes, achieving a precision of 0.92, recall of 0.88, and F1 score of 0.90. The findings highlight the potential of deep learning–based unsupervised methods to support real time, edge deployable solutions for energy efficiency and early fault detection in residential environments.]]>
Copyrights © 2025
