Research On Privacy Protection Scheme Of Internet Of Vehicles Based On Dummy Trajectory

The widely interactive location information in the Internet of Vehicles poses a great threat to the user’s trajectory privacy security.With the development of technologies related to Internet of Vehicles,this issue has received more and more attention.The dummy trajectory method is widely used to protect trajectory privacy because it does not require a trusted third party and can completely retain the user’s trajectory and many other advantages.However,due to the complexity of human behavior and the actual traffic environment,it is difficult for many current researches to fully learn and generate more realistic trajectories.In order to solve this problem,this thesis studies the behavior patterns of users in the continuous query and trajectory release scenarios of the Internet of Vehicles,and proposes a corresponding dummy trajectory privacy protection scheme.The main research works are as follows:(1)For the continuous query scenario,a prediction-based dummy trajectory privacy protection scheme is proposed.The basic idea of this scheme is to use the transition probability model to learn the users’ behavioral characteristics,and then generate dummy trajectories that conform to users’ moving tendency.First,by time slicing,many fine-grained historical query probabilities and a transition probability model are obtained.Then,two trajectory prediction algorithms are designed based on the transition probability model.After that,according to these algorithms and a dummy position generation algorithm,many dummy trajectories with the same direction as the real ones are generated.Finally,a query request is initiated using the anonymized set of all positions in the true and dummy trajectories,and the prediction results of the real positions are stored locally.When there is a query requirement,the user first searches in the local cache,and then performs the above process if the cache fails,thereby reducing the interaction between the user and the outside world.(2)For the trajectory release scenario,a dummy trajectory privacy protection scheme based on double Generative Adversarial Networks(GAN)is proposed.The basic idea of this scheme is to capture the intrinsic characteristics of trajectories of different lengths through a deep learning model that can realize mutual learning,and then generate dummy trajectories that conform to user behavior habits.First of all,double GANs are constructed based on a Recurrent Neural Network and the Informer,which is an improved version of the Transformer model.Then,a specified number of dummy trajectories are generated using double GANs.Afterwards,a road network matching algorithm is used to adjust generated dummy trajectoryies so that these trajectories highly matches the road network environment.Finally,a personalized k-anonymity mechanism is introduced to further optimize the scheme.Users can flexibly choose the level of privacy protection according to their own needs,and customize their own privacy protection scheme.(3)The feasibility and effectiveness of these two schemes are verified by theoretical security analysis and experimental comparison on real data sets.Experimental results show that both schemes can resist common malicious attacks in corresponding scenarios.Compared with the existing advanced schemes,the average passing rate of dummy positions of the first scheme is increased by about 30%,and the trajectory leakage rate of the second scheme is reduced by about 44%.It can be shown that these two schemes have better privacy protection effects in the corresponding scenarios,and can effectively protect the user’s trajectory privacy in the Internet of Vehicles.