Detection Method for False Data Injection Attacks Based on STCN-BiLSTM

To address the difficulties in feature extraction from long time-series data and the lack of interpretability in deep learning-based FDIA detection schemes, this paper proposes an FDIA detection method based on Spatiotemporal Convolution-Bidirectional Long Short-Term Memory (STCN-BiLSTM) networks. To prevent the loss of sequential information, the data in the dataset is first divided into two streams to extract their temporal features, and then the spatial features of the two streams are fused for learning, followed by classification and identification. Testing on the IEEE 14-bus system shows that the proposed scheme achieves a higher F1 score. Consequently, this method is more effective in detecting false data injection attacks.