Research On Security Policy Deployment And Privacy Protection At The Edge Of VANETs

With the gradual sinking of core network,new computing paradigm such as edge computing of VANETs has become a hot research topic.Because of the mobility,real-time and resource limited characteristics,traditional network security mechanism are difficult to apply.Only considering the mobility of existing technology and the combination of edge characteristics of the vehicle network with machine learning,and the establishment of active defense models such as edge intrusion detection of VANETs can the safety of VANETs services be guaranteed;under the mode of "intelligent model is service",AI privacy is more and more concerned,and the attack modes such as reverse attack emerge endlessly.This paper designs the anomaly detection scheme of VANETs by using HMM,explores the structure of the edge network and proposes the related model;the related model is improved to differential privacy version to protect model privacy;and the adaptive allocation algorithm of privacy budget based on gradient iteration algorithm is proposed.The main contents are as follows:The distributed IDS network edge security scheme is designed.HMM is used to detect unknown attacks at the vehicle cluster-head nodes,and all cluster-head monitoring nodes cooperate with each other to detect the outliers.Based on the security scheme,a mixture of HMM model(MHMM)is proposed to describe the distribution of network sequence data.The MHMM is improved by using the security similarity of the nodes brought by edge offloading.Combined with the vehicle movement in the region,the dynamic computing offloading MHMM model is obtained.The distance of offloading of VM nodes is calculated by parallel neural network.Baum welch algorithm and random gradient training are used to train HMM model parameters to detect the anomaly.The performance of the two models is tested on CIC IDS-2017 data set.The ROC curve and AUC are obtained for mixed attack mode.The influence of node similarity caused by computing offloading on the model is measured.An adaptive privacy budget allocation scheme is proposed,that is,an adaptive differential privacy mechanism ?-???? is proposed to avoid too much noise when the number of iterations is too large or close to the optimal value.Several privacy budget allocation mechanisms are proposed to solve the contradiction between privacy surplus and premature convergence.The simulation results show that the gradient can not only keep the confusion effect,but also keep the direction in the "main channel".