Deep Learning Model for Privacy Preservation of Vehicle Trajectories Over Internet of Vehicles

This research presents a deep learning-based differential privacy Laplace mechanism (DLPM) for the networking trajectories of the Internet of Vehicles. The DLPM is constructed using deep learning and clustering techniques to address consumers’ privacy leakage concerns effectively. Segment the trajectory space into separate regions based on timestamps to incorporate the temporal element in trajectories. This will determine the trajectory’s distribution points inside each zone. Enhance membership stability in multi-peak clustering for each region and pre-allocate the privacy budget matrix based on the trajectory density of each area. Employ a temporal graph convolutional network model to train and predict the designated privacy budget matrix while extracting the spatiotemporal characteristics of trajectory data. Disseminate the trajectory data just after applying Laplace noise to the predictive outcomes. Both theoretical and empirical research indicate that DLPM exhibits reduced overhead and more precise privacy budget predictions than alternative systems. The Differential Privacy Laplace Mechanism (DLPM) incorporates Laplace noise into trajectory data, enhancing its use.