A Research On Key Privacy-Preserving Technologies Of Mobile User Trajectory Data

With the development of mobile communication technology and intelligent devices,mobile users have generated mass trajectory data.User trajectory data contains rich spatio-temporal information.Analyzing and mining user mobility patterns can support various research and applications related to mobile trajectories,such as public governance and life services.However,trajectory data contains users’sensitive personal information.Improper trajectory data sharing will lead to the disclosure of user privacy,affecting the users’ life,property and personal safety,or even threatening political and public security.Therefore,society urgently demands the privacy release technology of trajectory data.This thesis mainly studies the trajectory privacy protection technology in the scenario of trajectory data publishing.It mainly researches the privacy protection of geographical trajectory location,trajectory trip characteristics and temporal correlation to provide high-intensity and measurable privacy protection for track data and ensure the availability of published data.The main contents are as follows:(1)To solve the problem of trajectory location privacy disclosure,this thesis proposes a trajectory location privacy protection method based on private grid division,which uses grid cells to index trajectory location points while ensuring that attackers cannot obtain user privacy through the grid structure.The grid structure is optimized by a hierarchical adaptive method to realize the separation of dense regions and the selection of partition scale based on density sensing mechanism to improve the accuracy of region query;Geographic constraints are proposed to limit the interference of excessive grid division on the geographic semantics of location points.Theoretical analysis and simulation results show that the proposed technology can effectively protect the track location privacy,and the generated data has good usability.(2)Aiming at the problem of privacy disclosure of trajectory travel features,this thesis proposes a privacy protection method of trajectory travel feature distribution based on differential privacy,which performs post-processing operations on the trajectory initial point distribution and initial segment distribution that meet the requirements of differential privacy,to improve the accuracy of generated data.This thesis constructs an optimal region merging and merges the noise counts of the adjacent geographic regions to offset some noise interference.Furthermore,it constructs a full k-ary tree of the initial segment and pruns it using the constraint relationship between nodes.The problem of sparse trajectory segments is solved,and private travel feature data availability is improved.Theoretical analysis and simulation experiments show that the proposed technology provides adequate privacy protection for the trajectory and ensures the accuracy of the generated data.(3)To solve the temporal correlation privacy disclosure problem,this thesis proposes a temporal correlation privacy protection method of trajectory based on the deep learning generation model,reconstructing the temporal correlation of the original trajectory data by the GRU(Gate Recurrent Unit).In order to effectively extract trajectory features,this thesis proposes a high availability trajectory generation model based on the GRU.The model takes the geographic semantic information of trajectory extracted by clustering algorithm as auxiliary information and the trajectory data as the input of GRU to mine the profound relationship between the spatio-temporal characteristics of the trajectory data.The simulation experiment evaluates the availability of the model and the generated data based on multiple metrics such as statistics and semantics.The experimental results show that the synthetic trajectory generated by the proposed method can better retain the features of trajectory data in point distribution,diameter distribution,frequent patterns and region query.(4)Aiming at the problem of trajectory temporal correlation privacy disclosure caused by the attack of the deep learning model,this thesis proposes a temporal correlation privacy protection method of trajectory based on the private depth generation model.The model optimization process is disturbed by the differential privacy mechanism to protect the privacy of the model and dataset.A model optimization perturbation algorithm based on Gaussian mechanism privacy protection is designed.By adding Gaussian noise to the model gradient,the model training process can meet the requirements of differential privacy.On this basis,this thesis introduces the zCDP(Zero-Concentrated Differential privacy)to improve the accuracy of the model by reducing the number of noise under the same privacy protection intensity.Theoretical analysis and simulation results show that the proposed technology can provide high-strength privacy protection for trajectory data and reduce the interference of privacy mechanisms on data availability.