Research On Industrial Control Intrusion Detection Technology Based On Gaussian Process And Hybrid Model

With the development of emerging technologies such as the Industrial Internet and 5G,there are more and more industrial control systems connected to the Internet,and the speed of industrial data transmission is getting faster and faster.However,related vulnerabilities in the Internet are also threatening the security of industrial control systems.Therefore,it is vital to ensure its security to avoid being attacked.However,due to the high dimensionality and redundant features of industrial control data,the existing intrusion detection technology still has many shortcomings.The key to the performance of intrusion detection is whether the appropriate classification algorithm is used and whether it is adjusted and optimized according to the characteristics of the data set.Therefore,based on the network traffic data generated by the industrial control system,this paper analyzes the characteristics of the industrial control data set,selects the most suitable characteristic value,and proposes the intrusion detection technology applied to the industrial control system.This article first proposes the CGWO-GP model,in which Gaussian process is a powerful and flexible probabilistic machine learning technology,suitable for processing data sets with high dimensionality,small samples,and nonlinear characteristics.The acquisition of its hyper-parameters is the key to restricting the performance of the Gaussian process.Therefore,this paper proposes an improved gray wolf optimization algorithm through analysis.The algorithm can jump out of the local optimal point many times by introducing the Cauchy mutation operator to expand the search of the Gaussian process.Optimize the scope to achieve the goal of global optimization.Finally,experiments show that the performance of the proposed algorithm has better results.Due to the applicable characteristics of the Gaussian process,the prediction of massive and unbalanced industrial control data sets has obvious defects.Therefore,this paper proposes the GMM-RNN model.During the operation of the industrial control system,its state is constantly changing,and a single Gaussian process cannot accurately describe the data set.Therefore,this article first takes advantage of GMM's ability to process unbalanced data sets to preprocess the original data.Secondly,the input data is classified and predicted by recurrent Neural Network.Recurrent neural network has a memory function and is especially suitable for data sets related to time series features.Finally,experiments were conducted on the UNSW-NB15 data set.The experimental results show that the GMM-RNN model performs better in the detection accuracy of attack types and real-time evaluation criteria,which significantly improves the performance of the system.