Interpretable Detection Method for False Data Injection Attack on Power Grid Based on Multi-head Graph Attention Network and Time Convolution Network Model

In the background of new power systems, fast and effective detection of false data injection attack (FDIA) is crucial for the safe operation of power grids. However, the existing deep learning methods do not fully explore the spatiotemporal feature information in measurement data of power grids, which affects the detection performance of models. Meanwhile, the“black box” attribute of deep neural networks reduces the interpretability of the detection model, leading to the lack of credibility in detection results. To solve the above problems, an interpretable FDIA detection method is proposed based on multi-head graph attention network and time convolution network (MGAT-TCN) model. First, considering the spatial correlation between power grid topology connection and measurement data, a spatial topology aware attention mechanism is introduced to establish the multi-head graph attention network (MGAT) to extract spatial features of measurement data. Next, the time convolution network (TCN) is used to extract the temporal features of the measurement data in parallel. Finally, the proposed MGAT-TCN model is simulated and validated on the IEEE 14-bus system and IEEE 39-bus system. The results indicate that the proposed model has higher detection accuracy and efficiency compared to the existing detection models, and the visualization of attention weights through topological heat maps can achieve interpretability of the model in spatial dimensions. © 2024 Automation of Electric Power Systems Press. All rights reserved.