A Novel Density Based Clustering Approach for Electricity Theft Detection

The extensive integration of advanced metering infrastructure (AMI) has transformed the landscape of electricity consumption monitoring, providing an effective avenue to address electricity theft by scrutinizing data obtained from smart meters. This research utilises the AMI data to detect instances of electricity theft by employing a novel unsupervised data-driven clustering technique called Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). The advantage of HDBSCAN in identifying points that do not conform to any cluster, commonly termed as noise points, becomes particularly pertinent in the context of theft detection. These noise points represent potential anomalies or irregularities in electricity consumption patterns, which are crucial for identifying and investigating suspicious activities. Leveraging HDBSCAN for outlier detection enhances the effectiveness of theft detection mechanisms by enabling the identification of data points that deviate significantly from the expected consumption patterns, thereby enhancing the security and integrity of electricity distribution networks. The proposed methodology requires consumption data from smart meters located at consumers premises only. The metering data undergo pre-processing, which encompasses tasks such as recovering missing and erroneous values, as well as normalization, ensuring the accuracy and reliability of the analysis. Furthermore, in our work, we have integrated Density Based Clustering Validation (DBCV) technique for the purpose of hyperparameter tuning, ensuring optimal performance and robustness of the proposed approach. The method was rigorously tested on two real-world datasets, and the results demonstrated robust performance across various evaluation metrics.