A Detection Based on OMES and MTAD-GAT for False Data Injection Attack in Smart Grid
- Resource Type
- Conference
- Authors
- Li, Huifeng; Li, Tiecheng; Chen, Tianying; Zhao, Guangxuan; Zhu, Yixiao; Kong, Xiangxing
- Source
- 2022 IEEE 6th Conference on Energy Internet and Energy System Integration (EI2) Energy Internet and Energy System Integration (EI2), 2022 IEEE 6th Conference on. :1578-1584 Nov, 2022
- Subject
- Power, Energy and Industry Applications
Support vector machines
Power measurement
Time series analysis
Energy measurement
Transforms
System integration
Time measurement
cyber-physical power system
false data injection attack
optimal measurement equipment scheduling
graph attention network
- Language
The false data injection attack (FDIA) is a new attack method aiming at power system state estimation. A successful FDIA can avoid the traditional bad data detection (BDD) and falsify measurements without detection. As a result, the dispatching center makes wrong decisions threatening safe and stable power system operation. A detection based on model-driven and data-driven methods is proposed. Optimal measurement equipment scheduling (OMES) is model-driven to calculate score 1 at each timestamp. Multivariate time-series anomaly detection via graph attention network (MTAD-GAT) is data-driven to get scores 2 and 3. The three scores above calculate the final inference score, which can maximize the overall effectiveness of FDIA detection. Timestamp can be identified as attacked if its final score is larger than a threshold chosen by Peak Over Threshold (POT) automatically. With the coordination of MTAD-GAT, OMES performs better detecting FDIA than the Support Vector Machine-Gabor Transform (SVM-GAB) method and Extreme Gradient Boosting (XG-Boost) method validated by simulations on the IEEE 14-bus system.