False Data Injection Anomaly Detection in Smart Grids: A Multi-Classifier OWA Data Fusion Approach

Abstract

Smart grids open up new opportunities through which a cyber intruder can infiltrate or manipulate data to compromise measurement integrity and state estimation accuracy. Advanced methods for detecting false data injection anomalies will be of great importance to the safety and reliability of power system operations. It therefore presents an Improved Threshold Prediction Anomaly FDIA Detection Approach that should finally address inherent limitations in traditional methods: limited adaptability to system changes, reduced sensitivity to complex anomalies, and incomplete coverage of emerging threats. The outputs of individual anomaly detectors are fused by utilizing an ordered weighted averaging fusion scheme in the method proposed herein. It improves sensitivity in detection and accuracy with significant countermeasures against FDIAs. In addition, Bayesian network-based hyperparameter optimization is utilized for each detector to refine them in a way that produces the best configuration towards maximum performance. Due to that, complementary strengths of the detectors provide a boost toward detection capability. Extensive experiments have been performed on real-world power grid data from NYISO using an IEEE 14-bus power system, and the robustness of the approach has been shown. Notably, at injection rates ranging from −20% to +20%, the proposed method demonstrated a 2.1% improvement in detection accuracy at +8% injection and a 10.7% improvement at −2% injection over the second-best state-of-the-art method. These results confirm the method's effectiveness in diagnosing and mitigating anomalies under a range of intrusion scenarios.