A Hybrid Model of GCN+GRU for False Data Injection Attack Detection in Smart Grid

False data injection attack (FDIA) can bypass the traditional bad data detection (BDD) and affect the operation of the power system. Considering the spatial-temporal characteristics of the smart grid, this paper proposes an FDIA detection model incorporating graph convolutional network (GCN) and gated recurrent unit (GRU). The data is decomposed into multiple wavelet coefficients by discrete wavelet transform (DWT) to reveal the different frequency components of the time series data. GCN applies convolution operation on graphic structure data to learn the spatial features between nodes. GRU effectively captures the dependencies in time series through its gating mechanism, which enhances the model’s ability to respond to dynamic changes in data to identify abnormal patterns more accurately. To verify the effectiveness of the proposed model, we simulate the IEEE-14 bus system and evaluate it through indicators such as accuracy, precision, recall rate, and $\mathbf{F 1}$-score. The simulation results show that compared with several traditional machine learning and deep learning algorithms, the hybrid model has better detection performance.