Research On Abnormal Traffic Detection Methods For Industrial Internet Based On Deep Learning With Time-space Fusion

In recent years,intelligent detection technology that is deeply integrated with the new generation of information technology is emerging.Industrial Internet has broken the closed production environment of traditional industrial control systems and strongly promoted the intelligent transformation of the industrial system,but it is facing severe information security threats.As an important network security protection technology,abnormal traffic detection can detect attacks in time by monitoring the network running status actively.However,the traditional abnormal traffic detection technology is difficult to extract effective features when faced with the current large and complex Industrial Internet traffic data,resulting in poor detection effect.This paper explores the Industrial Internet abnormal traffic detection methods based on the deep learning with time-space fusion.The main works and achievements are as follows:1.Aiming at the problem that the conventional abnormal traffic detection methods are hard to extract effective features from a great deal of complex Industrial Internet traffic data,resulting in low detection accuracy and F1 scorel,an abnormal traffic detection method for Industrial Internet based on the deep learning with time-space fusion is proposed.Firstly,the Industrial Internet traffic data are preprocessed to provide higherquality source data for the abnormal traffic detection model.Then,the deep learning model fused by the aggregated residual transformation network and the gated recurrent unit is employed to extract the spatial and temporal features of the traffic data and then obtain the final detection result by analysis.The test results indicate that the putted forward method can gain higher accuracy and F1 score than the comparison methods,reaching 95.45% and 89.21% respectively on the CSE-CIC-IDS2018 test set,and 95.65%and 85.65% on the Gas pipeline test set.Not only that,the proposed method has high detection efficiency.2.Aiming at the poor detection performance of Infiltration and NMRI samples,an abnormal traffic detection method for Industrial Internet based on unbalanced data is proposed.That is,the minority class abnormal traffic enhancement method based on variational autoencoder is applied to augment the Infiltration and NMRI class samples in the training set,prompting to balance the number of normal samples and the minority class samples that are difficult to be detected correctly in the training set.And then the deep learning model with time-space fusion is trained using the expanded training set.The experimental results reveal that the ameliorated method can not only keep the detection efficiency of the original method,but also validly exalt the detection performance of the fusion model for Infiltration and NMRI samples.Under the experimental conditions,the detection accuracy increased by 3.42% and 3.92%respectively,and the F1 score increased by 3.46% and 7.24% respectively.