Research On Intrusion Detection Technology Of Industrial Control Network Based On Deep Learning

Industrial intrusion detection refers to the monitoring and detection of industrial network traffic using computer technology to identify and prevent network intrusion attacks.With the rapid development of the industrial internet,the volume of data in industrial networks continues to expand,and network attacks have become more covert and sophisticated.Traditional methods are no longer applicable to the new scenarios of network intrusion detection.Deep learning-based algorithms for industrial intrusion detection offer effective solutions to the challenges faced by conventional techniques in handling large-scale,diverse,and irregular industrial network traffic data.The primary focus of this study was to validate and propose two deep learning-based models for industrial intrusion detection:the 1DLA-CNN model based on one-dimensional multiscale attention memory network and the DCNN-IDS model,an intrusion detection algorithm based on deep convolutional neural network.The specific tasks carried out are as follows:(1)The validation of an intrusion detection algorithm model based on 1DCNN-LSTM was conducted in this article.Through experiments conducted on the natural gas dataset from Mississippi State University and the CICIDS2017 dataset,the effectiveness of the intrusion detection algorithm model based on 1DCNN-LSTM was confirmed.The experimental results demonstrated outstanding performance of the model on the natural gas dataset,achieving an impressive accuracy of 95.56%.Although the model showed suboptimal performance for attack types such as Botnet,DoS,and Port on the CICIDS2017 dataset,the overall accuracy still reached 93.01%.(2)In response to the issues of weak feature extraction capability and poor generalization of the 1 DCNN-LSTM model,this study proposes an improved version called the 1DLA-CNN model.This model addresses the limitations by introducing multiscale convolution and attention mechanism,which allows for the fusion of information from different scales,thereby enhancing the feature extraction capability and optimizing accuracy and performance.Experimental results demonstrate that the enhanced 1DLA-CNN model performs exceptionally well on both datasets.On the natural gas dataset,it achieves an accuracy of 98.23%,while on the CICIDS2017 dataset,the accuracy reaches 95.01%.(3)To address the challenges of feature selection and overfitting in traditional machine learning approaches when dealing with industrial bus data,this paper proposes the DCNN-IDS model,which is validated using the CAN bus dataset.The DCNN-IDS model employs the Inception-ResNet model as its foundational structure,optimizing the model size and redesigning the network architecture.Additionally,it introduces an image generator module to generate two-dimensional CAN data images with sequentially ordered bit identifiers.Experimental results demonstrate the outstanding performance of the DCNN-IDS model,achieving significant effectiveness in detecting intrusion attacks in the CAN bus dataset,particularly in complex scenarios involving irregular and random attacks.In conclusion,deep learning-based intrusion detection algorithm models demonstrate excellent performance in industrial network security,effectively detecting intrusion attacks and improving network security performance.The proposed model in this study achieved higher accuracy and better generalization capability in experiments,effectively identifying intrusion behaviors in industrial network traffic data.This provides strong support for the development of industrial network security field.