Research On Abnormal Traffic Detection Of Industrial Control Based On Neural Network

With the industrial control systems deeply integrated to networks,the industrial control network greatly simplifies the work of the industrial control industry.But at the same time,the increasingly serious network security problems also cast a shadow over its promotion.Abnormal traffic detection is a common means to maintain network security by collecting and analyzing network traffic to identify abnormal traffic,which would timely respond to possible attacks,and protecting network security.With its unique convolution operation,the convolutional neural network can effectively extract feature information and distinguish complex data.With its application in abnormal traffic detection,the system's ability to feature analysis and anomaly identification can be improved.Therefore,based on the convolution neural network,this thesis optimizes the network structure and proposes an abnormal traffic detection model.This thesis first analyzes the data imbalance problem that often exists in anomaly detection data,then describes the causes,consequences,and different solutions to the problem in detail.In this thesis,we choose the combined sampling method to balance the data.Firstly,we use the improved SMOTE method to expand the data,so that the minor types of samples are increased.For the problem of sample space overlapping,in order to reduce the impact of noise,Tomek Link undersampling method is adopted to sanitize the data.The model based on the convolution neural network is used to detect abnormal traffic.With its unique convolution operation,the convolutional neural network can extract valid feature information from the parts to the whole.Furthermore,the spatial structure and feature weight of the network are further optimized,and a multi-scale skipping excitation junction is proposed at the same time.Based on the characteristics of abnormal traffic data,several modules in the network are also optimized.In this thesis,the proposed model achieves an accuracy rate of 93.67% on the KDD cup 99 data set,which can effectively detect abnormal traffic.