DAMGAT-Based Interpretable Detection of False Data Injection Attacks in Smart Grids

False data injection attacks (FDIAs) significantly disrupt the secure operation of smart grids by manipulating the measured values collected by intelligent instruments. Existing studies have utilized deep learning techniques to enhance the detection of FDIAs, however, these studies often overlook the spatial correlation between power grid topology and measurement data. Meanwhile, the high complexity of deep neural network severely impedes the interpretability of detection models, resulting in the incredibility of detection results. To address the above challenges, this paper proposes an interpretable deep learning FDIAs detection method, named dual-attention multi-head graph attention network, DAMGAT. The DAMGAT introduces a dual-attention mechanism that incorporates both node feature attention and spatial topology attention into a multi-head graph attention network. This mechanism efficiently aggregates attack characteristics and spatial topology information by dynamically capturing the potential correlations between FDIAs detection and measurement data. Furthermore, the proposed model can provide clear and credible interpretations for high-accuracy detection results via analyzing features and spatial topology attention weights. Extensive simulations are performed using the IEEE 14-bus and 118-bus test systems. The experimental results demonstrate that the proposed model outperforms state-of-the-art FDIA detection methods in terms of accuracy, while also providing reasonable interpretability for features and spatial dimensions.