Research On Edge Network Intrusion Detection Based On Machine Learning

With the advent of the information age and the Internet of Everything,China’s Internet usage rate has increased year by year,and its development and innovation have been accelerating.In order to solve the pressure of a large amount of data computing on the cloud service center,the concept of edge computing came into being.Edge computing is to transfer some data aggregation,storage and analysis tasks to edge computing nodes closer to users,so as to reduce the computing pressure of cloud center and improve network service efficiency.However,edge nodes need to face more diverse network scenarios and more complex network environments,so the probability of edge networks being threatened by security is higher than that of ordinary networks.At the same time,edge nodes have limited resources,so they are easy to become the target of network intruders,and the probability of collapse caused by network attacks is also higher.Therefore,real-time awareness of edge network situation and timely and effective detection of network intrusion threats are carried out,It is crucial for edge networks to provide continuous and stable services.In view of the above situation,this article studies the intrusion detection of edge networks based on machine learning methods,and proposes an intrusion detection model based on feature union network and an intrusion detection model based on multi center incremental clustering for the detection of known attacks and unknown attacks of edge networks,including the following contents:In view of the complex network environment of edge networks and the numerous network threats faced by them,this paper proposes an intrusion detection model based on feature joint networks by considering the contribution of the relationship between features to the network intrusion detection results.The improved Text CNN and Factorization Machines are used to mine data features from the deep and wide dimensions,extract the high-order features and low order features of the samples,and associate the two parts of feature information to detect network intrusion;The second order combination of features is used to consider the gain effect between features in network attack samples and reduce the impact of sparse samples on the detection results.Aiming at the feature that the attack forms of edge network are iterative and new attack forms emerge in endlessly,an intrusion detection model based on multi center incremental clustering algorithm is proposed.The model improves the DBSCAN algorithm,and uses the existing classification tags to make clustering results more accurate;By introducing the concept of multi class cluster center,we describe the characteristics of clusters with large number of samples and irregular distribution through multiple class cluster centers;At the same time,for the newly captured incremental data samples,the class cluster center around the sample is used to judge the category of incremental data,which improves the accuracy of model intrusion detection and reduces the detection time for new data.