Hybrid Anomaly Detection Framework for Electricity Theft Identification in Smart Grid Systems

This paper presented a Hybrid AI-Based Electricity Theft and Anomaly Detection System addressing the dual challenges of detection accuracy and operational transparency in smart grid environments. By integrating XGBoost for supervised classification, Isolation Forest for unsupervised anomaly iso-lation, and CNN-1D for temporal pattern learning within a stacking ensemble, the proposed architecture achieves strong and consistent anomaly detection performance. XGBoost de-livers perfect recall of 1.00 with 99% accuracy, ensuring no theft event is overlooked [6], while CNN-1D captures complex sequential consumption patterns that shallow models cannot represent [7]. The confusion matrix analysis (Fig. 4) confirms that the Hybrid ensemble provides the most operationally balanced outcome among non-XGBoost configurations, and the model performance comparison (Fig. 3) highlights the complementary metric profiles that motivate the stacking de-sign [10], [16]. A SHAP-based explainability module [11] accompanies every alert with human-readable feature attribution, addressing the interpretability gap that limits adoption of purely black-box AI systems [18]. The framework is deployed through an interactive Streamlit dashboard supporting real-time visuali-sation, batch review, and investigation reporting, making the system accessible to utility administrators without requiring data science expertise. Future work will investigate federated learning strategies [20] for training across multiple utility networks without centralising sensitive data. Incorporating graph-based anomaly detection to model spatial relationships within distribution net-work topology, and exploring adaptive reinforcement learning approaches [19] to respond to novel theft strategies, represent further promising directions for improving detection coverage.

[1]     Northeast Group LLC, “Global electricity theft: Quantifying the chal-lenge,” Northeast Group LLC, Washington D.C., USA, Tech. Rep., 2022. [2]     International Energy Agency (IEA), “Electricity security and non-technical losses,” International Energy Agency, Paris, France, Tech. Rep., 2023. [3]     R. Jiang, R. Lu, Y. Wang, J. Luo, C. Shen, and X. Shen, “Energy-theft detection issues for advanced metering infrastructure in smart grid,” Tsinghua Science and Technology, vol. 19, no. 2, pp. 105–120, Apr. 2014. [4]     G. M. Messinis and N. D. Hatziargyriou, “Review of non-technical loss detection methods,” Electric Power Systems Research, vol. 158, pp. 250–266, May 2018. [5]     J. Nagi, K. S. Yap, S. K. Tiong, S. K. Ahmed, and M. Mohamad, “Nontechnical loss detection for metered customers in power utility using support vector machines,” IEEE Transactions on Power Delivery, vol. 25, no. 2, pp. 1162–1171, Apr. 2010. [6]     Z. Yan and H. Wen, “Electricity theft detection base on extreme gradient boosting in AMI,” IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–9, 2021. [7]     Z. Zheng, Y. Yang, X. Niu, H.-N. Dai, and Y. Zhou, “Wide and deep convolutional neural networks for electricity-theft detection to secure smart grids,” IEEE Transactions on Industrial Informatics, vol. 14, no. 4, pp. 1606–1615, Apr. 2018. [8]     L. J. Lepolesa, S. Achari, and L. Cheng, “Electricity theft detection in smart grids based on deep neural network,” IEEE Access, vol. 10, pp. 39 638–39 655, 2022. [9]     F. T. Liu, K. M. Ting, and Z.-H. Zhou, “Isolation forest,” in Proceedings of the IEEE International Conference on Data Mining (ICDM), Pisa, Italy, Dec. 2008, pp. 413–422. [10]    S. K. Gunturi and D. Sarkar, “Ensemble machine learning models for the detection of energy theft,” Electric Power Systems Research, vol. 192, p. 106904, 2021.