Learning long- and short-term dependencies for network intrusion detection

Accurate network intrusion detection is very important for the security of power information system. In this study, we introduce an innovative architecture designed for network intrusion detection to enhance the security of power grid information systems. Our proposed framework integrates the WaveNet and BiLSTM models to effectively capture both short-term and long-term sequence dependencies present in intrusion traffic. The WaveNet model is employed to capture short-term dependencies by effectively modeling local patterns. In contrast, the BiLSTM model is utilized to capture long-term dependencies by excelling in capturing broader patterns across extended sequences. This combination of models enables a comprehensive understanding of the complex temporal dependencies exhibited in intrusion traffic data. To further enhance the model's performance, hybrid pooling layers are incorporated, comprising both maximum pooling and average pooling layers. This combination enables the capture of both global and local features, resulting in a more comprehensive data representation. The proposed model is rigorously evaluated using multiple datasets, demonstrating competitive intrusion recognition accuracy. These results emphasize the model's effectiveness in safeguarding the security of power grid information systems.