| Abstract:With the large-scale integration of smart terminals in distribution networks, grid intelligence and productivity have been greatly improved, but the risk of cyber intrusion has also increased. The wide integration of the terminals leads to fragmented attack sources, diverse intrusion types, and high randomness, which challenge conventional intrusion detection models in accurately identifying threats and meeting grid security demands. To address this problem, an AG-CWGAN-TCN-GCN based intrusion detection model for integrated distribution terminals is proposed in the paper. First, an attention-guided Wasserstein GAN is designed to generate intrusion samples, mitigating class imbalance caused by scarcity. Second, a TCN-GCN framework is constructed to capture spatio-temporal features from network traffic data, dynamically filtering key attack patterns from high-dimensional flow characteristics, and improving feature extraction efficiency. Finally, a cross-attention mechanism is incorporated to enhance deep interaction and weight allocation of critical features, enabling fuller representation and fusion of intrusion traits. Experimental results show that the proposed model outperforms classical intrusion detection methods in key metrics such as accuracy, detection rate, and F1-score, demonstrating strong detection capability and practical utility. |
点此下载全文
位访问者,今日访问
