TLPP: Deep-Learning-Based Two-Layer Privacy Preserving Mechanism for Protecting Vehicle Trajectory Data

With the popularity of the global positioning system (GPS) and mobile Internet, a large amount of vehicle trajectory data has been generated and applied in intelligent transportation systems. The collected trajectory data often contains sensitive user information, which poses a risk of user privacy disclosure. To enhance the privacy of vehicle trajectory data, this article proposes a novel two-layer privacy preserving (TLPP) mechanism that leverages clustering features. Initially, density-based clustering is employed to derive regional attributes and density characteristics of trajectory points. Subsequently, a generative adversarial network (GAN) incorporating a long short-term memory (LSTM) network is utilized to learn the distribution of clustered trajectories, facilitating the generation of synthetic trajectories. These synthetic trajectories are then substituted for the original trajectories, constituting the first layer of privacy protection. To ensure the fidelity of the synthetic data, a novel generator loss function is designed, utilizing the Wasserstein distance to quantify the spatial similarity between the real and synthetic trajectories. Furthermore, to accommodate the personalized privacy requirements, a tailored differential privacy mechanism is introduced. This mechanism provides a second layer of privacy protection by introducing the region-specific perturbations to the data. The experimental results show that, compared with the other models, our approach can effectively protect the user privacy while ensuring the trajectory data utility.