Industrial Intrusion Detection System Based on CNN-Attention -BILSTM Network
Huang Chi, Lin Chen · 2022
In view of the existing intrusion detection system for real-time control, the problem of accuracy is not high, this algorithm with improved genetic algorithm to optimize the input vector of belief network, after dimension convolution layer is utilized to extract local features, recycling attention mechanism to explore the characteristics of the different impact on the weights of attack types of prediction, finally USES two-way LSTM extraction sequence. In this paper, the model based on CNN-attention-BILSTM network is compared with the model based on convolutional neural network and attention-BILSTM network and other technologies, and the performance is evaluated from the perspectives of accuracy, false alarm rate, processing performance and completeness. Finally, the experimental results show that, compared with the other models, the model based on CNN-attention-BILSTM network has the best performance on the KDD 99 test set. Using belief network dimension reduction can effectively reduce program running time.