Research And Implementation Of Trusted Industrial Internet Intrusion Detection Model

The arrival of 5G era has triggered a new round of industrial reform.The rapid development of the Internet has realized the interworking and resource sharing of industrial manufacturing industries,and promoted the development of industrial manufacturing towards a more intelligent direction.However,while promoting the industrial development,there are also various threats of network attacks.At present,there are frequent attacks on the industrial manufacturing industry around the world.Therefore,the application of intrusion detection technology to the industrial Internet has become a research hotspot.Artificial intelligence has developed rapidly in recent years.Researchers widely use machine learning and deep learning to detect and analyze network traffic.When processing large-scale data,machine learning algorithm has its limitations,requiring artificial design of feature extraction methods,which seriously affects the accuracy of final prediction results.While deep learning algorithm can automatically learn category features from a large number of data samples,saving labor costs and ensuring the accuracy of prediction.However,the industrial Internet environment is complex.There are various attacks and a lot of noise in traffic data,which makes general deep learning algorithms unable to accurately learn important features in data.Meanwhile,deep learning algorithms have opaque defects.Highly complex models make deep learning into a black box,which makes it impossible for us to reasonably explain the final predicted results.The lack of interpretability also severely hinders the further development of deep learning.In response to the above problems,this paper designs and realizes a CMAG model integrating convolutional neural network(CNN),Multi-head Attention mechanism and bidirectional gated circulation unit(Bi GRU)to fully learn important features of network traffic,so as to improve the accuracy of the model.At the same time,the current popular interpretability method SHAP is adopted,and the Chi-square test method is adopted to improve the problem of SHAP equalizing feature contribution in the case of multi-feature interaction.Then,the Chi-square SHAP is used to explain the CMAG model,and the key features in the prediction process are visualized to improve the transparency of the model.The main contributions of this paper are as follows:(1)This paper designs and implements a new intrusion detection analysis model CMAG,which integrates three deep learning methods,and conducts experimental comparison with some existing intrusion detection models CNN,CNN+GRU and CNN+Bi GRU on the CSE-CIC-IDS2018 dataset.Experiments show that CMAG model has higher accuracy and better performance.(2)In this paper,the Chi-square test method is adopted to improve the SHAP explicable method,and the problem of averaging the contribution of features under the combined action of multiple features is improved.Then,the Chi-square-Shap method is used to explain the designed deep learning model and improve the transparency of the model.Experiments show that the Chi-square-SHAP method can explain the main characteristics of various attacks more effectively.(3)This paper builds a credible industrial Internet intrusion detection system.The system uses the CMAG deep learning model and the explainable method of Chi-square-Shap designed in this paper,and presents it to users in a visual way to increase the explainability of the system and enhance users’ trust.