Research On Intrusion Detection Method Of Industrial Control System Based On SVM

Industrial Control System is widely used in electric power,transportation,chemical industry and other industries.It is an important part of the national key infrastructure.With the connection of industrial control systems with the outside world and the use of common network protocols,the security threats to industrial control systems are also increasing.In response to this problem,scholars at home and abroad are studying the safety protection of industrial control systems.Among many security protection technologies,intrusion detection technology can effectively detect attack behaviors,improve the security performance of industrial control systems,and identify external attacks.Based on reading a large number of domestic and foreign research literatures,this paper summarizes the composition structure of ICS,the weak links of the system and the intrusion detection technology.Based on the research results,the network security features of ICS are analyzed,and the network data of ICS is used as the data source for intrusion detection,and the following work is completed:(1)A feature dimensionality reduction method combining Filter-Wrapper and KPCA is designed.For ICS network data,which has high dimensionality,nonlinearity,multiple redundant features,and strong correlation between features,it combines the feature selection method of Filter and Wrapper In order to find the optimal feature subset from the original data features,the KPCA method is used to map the nonlinear data into the high-dimensional space,and the low-dimensional mapping of the data features is realized.(2)Introduce the stacked auto-encoder in deep learning into the feature dimensionality reduction of ICS network data,establish a multi-layer neural network model,perform SAE training by layer-by-layer training and mini-batch-size methods The output of is used as the input data of SVM,and the accuracy of the model is used as the evaluation standard to determine the optimal dimension reduction dimension of SAE.(3)For the optimization of the parameters of the SVM intrusion detection model,on the basis of the standard particle swarm algorithm,an adaptive mutation particle swarm algorithm SVPSO is proposed.Based on the feature dimensionality reduction,the SVPSO algorithm is used to perform SVM The optimization of the penalty factor C and the parameters of the kernel function parameters.(4)The overall design of the SVM intrusion detection model.Based on the feature reduction of ICS network data and the optimization of SVM parameters,this paper establishes the SVPSO-SVM intrusion detection model combining Filter-Wrapper and KPCA and the SVPSO-SVM intrusion detection model based on SAE in the above method.Based on the ICS intrusion detection data set proposed by Mississippi State University,the experimental results show that the detection accuracy of the SVPSO-SVM model based on SAE reaches 95.75%,and the detection accuracy of the SVPSO-SVM model combining Filter-Wrapper and KPCA reaches 98.625%,compared with other methods,the method proposed in this paper can obtain higher detection accuracy with lower feature dimension.