Network Intrusion Detection Model with Multi-Layer Bi-Directional Temporal Convolutional Networks and Multi-Headed Self-Attention Mechanisms

Various attacks under current large-scale network traffic have caused adverse effects, and intrusion detection against malicious traffic has become increasingly important.However, traditional intrusion detection methods based on misuse and machine learning suffer from the problems of relying on human experience and low detection efficiency, Based on this this paper proposes a novel and improved multilayer BiTCN-MHSA model:Eff-MHSA.In the Eff-MHSA model, firstly, the preprocessed network traffic is generated into a bi-directional sequence by utilizing the bi-directional sliding-window method,Bidirectional Temporal Convolutional Nework (BiTCN) Multi-Layer Network. BiTCN (BiTCN) multi head self MHSA (MHSA) mechanism is fused into the MATH module, the bidirectional temporal features are extracted and fused in an additive way to obtain the fusion features after the temporal information is enhanced. Different weights are assigned to different subsequences of network traffic to extract key features. The EQL v2 loss function is used to solve the data class imbalance problem. The Eff-MHSA is evaluated on the benchmark network attack dataset NSL-KDD. the model performances are all better than traditional algorithms and outperform other deep learning models.