An improved iForest-based method for electricity consumption anomaly detection

The widespread application of artificial intelligence and intelligent algorithms in large-scale time-series data analysis has provided new technical approaches for detecting abnormal electricity consumption in power systems. Traditional anomaly detection methods often suffer from insufficient recognition accuracy and stability in practical applications due to the characteristics of electricity consumption data, including pronounced periodicity, complex fluctuation patterns, and a low proportion of abnormal samples. To address these challenges, this paper proposes an improved iForest-based method for electricity consumption anomaly detection, enhancing the model's ability to recognize diverse anomaly behaviors. Building upon the Isolation Forest framework, the method optimizes feature representation and refines the anomaly scoring mechanism, enabling more effective characterization of local fluctuations and continuous anomalies within time-series consumption data. Additionally, a comprehensive detection workflow encompassing data preprocessing, anomaly detection, and alert analysis is established. Experimental results demonstrate that on real electricity consumption datasets, the proposed method outperforms traditional iForest and other comparative models in anomaly discrimination, detection stability, and comprehensive performance metrics. It achieves more reliable anomaly identification while maintaining controllable computational complexity. Findings indicate that the proposed method provides effective technical support for electricity anomaly monitoring and operational management, exhibiting strong potential for engineering applications.