Anomaly detection method for power dispatch streaming data based on adaptive isolation forest and self-supervised learning

Introduction To address the issues of concept drift and scarcity of anomaly samples in real-time anomaly detection under the massive streaming data environment of power dispatching and control systems, this study focuses on developing an effective detection method. Methods We propose a streaming data anomaly detection method integrating adaptive isolation forests and self-supervised learning. First, a business-based model is constructed by analyzing inherent relationships between system services, processes, and resource usage. An improved isolation forest algorithm with a sub-forest progressive update mechanism is designed—selectively eliminating sub-detectors with large anomaly rate deviations and dynamically adding new ones to overcome performance degradation from traditional random updates. Additionally, a GPT-based self-supervised learning framework is introduced, incorporating state memory units to encode historical data patterns and a distance metric-based sampling strategy to reduce redundancy. Results Experiments on a real power dispatching process resource dataset show the proposed method significantly outperforms the traditional streaming data isolation forest algorithm in key indicators such as AUC value, with the highest improvement reaching 39.12%. Ablation experiments verify the effectiveness of each module. Discussion The proposed method enhances the detection algorithm's adaptability to concept drift, overall stability, and ability to perceive hidden anomalies, providing reliable technical support for the safe and stable operation of the power dispatching system.