Research On Deep Learning Based Intrusion Detection Method For Industrial Control Network

Industrial control systems(ICS)such as natural gas,sewage treatment plants,and smart grids are key national infrastructures.Once their networks are attacked,they will not only cause huge property losses,but also endanger human health.Therefore,the security of industrial control networks have drawn much attention.Due to physical isolation of traditional ICS,its industrial control network is in a safe and closed state.However,with the combination of industrial automation and information technology,more and more industrial control systems are connected to the Internet,which improves work efficiency and also leads to industrial control.Networks are vulnerable to intrusion threats,so effective methods are urgently needed to ensure the safety of industrial control system network.This article mainly focuses on the research of intrusion detection methods for industrial control system network,including the following research contents:(1)Research on industrial control system network anomaly detection based on bidirectional generative adversarial network(Bi GAN).To solve the problem of long test time of anomaly detection model based on generative admissibility network,the flexible reconstruction mechanism of Bi GAN is introduced to establish anomaly detection model to improve the real-time performance.At the same time,LSTM is incorporated into Bi GAN generators,discriminators and encoders to strengthen the Feature learning ability of industrial control network data.In addition,the objective function of parameter optimization was established by introducing weights,and the accuracy of verification set was taken as the index of updating parameters to complete the hyperparameter optimization,so as to improve the detection performance of the model.A comparative experiment on the industrial control system network standard data set shows that the proposed algorithm has a better detection effect.(2)Research on misuse detection of industrial control system network based on hybrid neural network and attention mechanism.To solve the problem of high false alarm rate in industrial control system network anomaly detection,a misuse detection scheme combining hybrid neural network with attention mechanism is proposed.In this scheme,convolutional neural network and bidirectional gated recurrent unit were used to extract data features from spatial dimension and temporal dimension respectively to enrich the description of feature information.In addition,on the basis of the hybrid neural network,the attention mechanism is combined to automatically give higher weight to the feature information that has a greater contribution rate to the detection result,so as to achieve the purpose of optimizing the feature vector and further improve the detection effect.Finally,a comparative experiment with other related detection models on the industrial control system network data set shows that the false alarm rate of the designed detection model has been significantly improved.(3)Research on the automatic optimization method of hyperparameters of industrial control system network intrusion detection model based on deep learning.In view of the existing industrial network intrusion detection models based on deep learning,and the problem that the hyperparameter optimization methods of anomaly detection models and misuse detection models in Research 1 and Research 2 are highly complex and difficult to obtain the optimal hyperparameter,a new industrial control system network intrusion detection strategy combining the automatic optimization algorithm of hyperparameter and stacked LSTM is proposed.This strategy uses the Bayesian optimization algorithm to search for the optimal hyperparameters of the stacked LSTM detection model,and realizes the automatic optimization of the hyperparameters of the industrial control system network intrusion detection model based on deep learning,and the proposed hyperparameter automatic optimization algorithm combined with detection model in Research 1 and Research 2,through comparative experiments,it is found that the proposed algorithm can further improve the performance of model detection and give full play to the potential of the model.