Research On Network Intrusion Detection Method Of Cyber-physics System

Cyber-Physical System(CPS)is an intelligent system formed by the organic integration of computing,communication and control.It is at the core of the "Industry 4.0" and "Made in China 2025" strategies.A typical CPS The system mainly includes sensors,actuators,controllers and a communication network.The communication network is used to complete the information interaction between system components and is the core of CPS.Therefore,CPS should pay more attention to network security.Nowadays,CPS has been widely used in fields closely related to people's lives,such as smart grids,transportation networks,and medical systems.How to ensure the safety of CPS has become the primary problem to be solved.Intrusion Detection System(IDS)has the characteristics of active protection,which can respond and deal with intrusions in CPS in a timely manner.Nowadays,the rapid development of 5G information technology has led to larger network traffic data.The detection performance of IDS based on traditional machine learning is often closely related to the feature extraction of data.Feature extraction requires manual processing and requires a large number of complex calculations.Strong dependence.Deep learning can automatically learn feature representations from data features,reducing the dependence on manual extraction of features.Therefore,this paper applies CNN to the IDS of CPS.In order to improve the generalization ability of the model,this paper uses the Dropblock technology in the convolutional layer to randomly inactivate the feature maps in the form of spatial blocks,which effectively solves the problem that the traditional Dropout method is not applicable in the convolutional layer.In the data set of network intrusion,the sample sizes of different attack types are quite different,resulting in uneven data distribution,which makes the model more inclined to categories with larger sample sizes.In order to solve this problem,the Focal Loss loss function is used in the model in this paper.Improve the model's ability to process unbalanced data samples,and reduce the impact of unbalanced sample sets on model performance.This paper improves the traditional CNN structure,combines the Inception structure and the dense convolutional network(Dense Net)model to propose an Inception?Dense Net block,so that the model can learn richer features in each layer,and the model can be better the feature information of the original data is retained,so that the output result of the model is not only dependent on the features of the last layer in the network,and the reusability of features is improved.And adjust the parameters of the model,such as activation function,optimization algorithm,etc.,to make the model optimal.This article uses the KDD CUP99 data set to test the performance of the model,and uses traditional machine learning models and classic CNN models to compare experiments with the model in this article on the same data set.Analyzing the results,it is concluded that the overall recognition ability of the model in this paper is improved compared with other models.Therefore,the usability of this model in CPS intrusion detection system is verified.CPS is real-time,so in order to improve the detection speed of the model,this article uses the Tensorflow deep learning framework to build a distributed intrusion detection system,and compares the training speed of the model in a stand-alone environment.The results show that the distributed environment can effectively improve the detection of the model.Speed,so that it can better meet the real-time nature of CPS.