Research On Deep Learning Method Of Intrusion Detection In Industrial Cyber-Physical System

The fourth industrial revolution promoted the rapid development of emerging technologies such as wireless communications,cloud computing,big data,artificial intelligence,and smart driving cars,and gave birth to the Industrial Cyber Physical System(ICPS).However,as open wireless communications gradually replace wired connections,ICPS faces both traditional and harsh industrial environment interference and various contemporary cyber security threats.This paper starts with constructing an effective intrusion detection model,with the goal of enhancing the system's ability to recognize attack behaviors in a massive data environment,and studies intrusion detection strategies based on deep learning in industrial cyber-physical systems.The main research contents are as follows:(1)Analyze the basic concepts,architectural methods,and research value of industrial cyber-physical systems,and tell which cyber attacks threaten industrial cyber-physical systems in the real production environment.Select the natural gas pipeline SCADA system under ICPS for research,and design the intrusion detection system architecture based on deep learning in this scenario.(2)Research on ICPS intrusion detection method based on deep learning hybrid model.The Deep Belief Networks(DBN)in the neural network algorithm is selected as the main body data dimensionality reduction part,and the classic machine learning algorithm Support Vector Machine(SVM)is used as the classification detection part.The preprocessed industrial traffic data set is imported as the visible layer of the dimensionality reduction part of the DBN;in the classification and detection,the SVM is used to classify and identify the dimensionality reduction data to determine whether the system has been maliciously attacked.The simulation experiment results show that the intrusion detection method proposed in this paper has both good recognition accuracy and detection efficiency.(3)Research on intrusion detection method of industrial cyber-physical system based on integrated learning of deep auto-encoding networks.The intrusion detection method of industrial cyber-physical system based on deep learning hybrid model cannot achieve multi-classification,and the use of DBN model as a data dimensionality reduction module cannot improve the anti-interference ability of the data,resulting in problems such as low generalization ability of the model.Therefore,the introduction of Deep De-noising Auto-Encoding Networks(DAN)and Integrated Deep Belief Networks(EDBNs)models to form an integrated learning industrial cyber-physical system intrusion detection based on deep self-encoding networks method.The preprocessed industrial flow data is passed into the DDAN module,and the reduced dimensionality data is output and imported into the integrated learning model EDBNs,and finally the prediction result is obtained.The simulation experiment results show that compared with the industrial cyber-physical system based on the deep learning hybrid model,the method proposed in this paper can overcome the shortcomings of SVM that can only achieve two classifications,and due to the introduction of the integrated learning model,it reduces the system's most important parameters for a single model.The excellent dependence also significantly reduces the time cost of manual tuning,and has higher practicability and recognition rate.(4)Research on intrusion detection method of industrial cyber-physical system based on data enhancement.Aiming at the intrusion detection method of the integrated learning industrial cyber-physical system based on the deep self-encoding network,the accuracy of identifying a specific type of traffic is low when the training samples are unevenly distributed,and the data enhancement module is introduced.First,the random forest algorithm is used to filter the input features,and the main features and non-main features are divided,and the samples composed of the main features are used as the samples of the training model;then,the unbalanced data filtering layer is used to find a certain category of samples with few training samples.Use Generative Adversarial Networks(GAN)to enhance the data of this class of samples to generate the number of training samples for this class;finally,use the DBN model combined with the softmax classifier to classify and recognize the enhanced data samples.The method proposed in this paper can effectively fit training samples,increase the number of training samples,and effectively improve the generalization ability of intrusion detection systems.