A Graph Neural Network and Transformer Encoder Technique for Anomaly and Cyber Threat Detection in Smart Grids

Nowadays, Smart grids enhance traditional power grids by integrating information and communication technologies for more efficient power delivery and management. However, this integration introduces security vulnerabilities and their exploitation leads to cyber-attacks causing severe consequences like infrastructural damage, massive blackouts. This paper presents an effective approach to detect anomaly and cyber threat in smart grids based on deep learning. A novel Graph Neural Network and Transformer encoder (GNN-Transformer encoder) feature extraction method is proposed in this paper for anomaly and cyber threat identification. Initially, the input data is normalized using min-max normalization technique to transform the data into a common scale. The GNN is utilized for spatial feature extraction which conducts representation learning based on topology of power system. The transformer encoder with self-attention mechanism determines the importance of different time steps and captures long-range dependencies. The fully connected layers provided accurate prediction results by mapping the high-dimensional feature vectors to the final output. The proposed method is evaluated on PMU dataset and achieved high accuracy 97.90%, precision 0.82, recall 0.98. The GNN-Transformer encoder method outperformed existing anomaly detection models such as Gaussian Mixture Model (GMM), Spatial Pyramid Pooling with Convolutional Neural Network (SPP-CNN).