Vehicle Trajectory Generation Based on Generation Adversarial Network

With the development of networked vehicles, location information-based transportation systems have proven to provide significant benefits. However, the exposure of vehicle location information also raises important privacy issues. Current typical methods for protecting vehicle location privacy protection methods such as anonymity and pseudonymity, still carry the risk of the vehicle being tracked, leading to data security issues. This paper proposes a kind of vehicle trajectory generation algorithm based on Generative Adversarial Networks (GAN). The algorithm utilizes vehicle movement trajectory data to train both the discriminator and generator models to generate virtual trajectory data that matches the distribution of real trajectory data. Therefore, virtual trajectory data can obscure vehicle information, addressing the privacy concerns associated with moving trajectory data and enhancing the security of applications. In this paper, the vehicle travel time of sample trajectory data and virtual trajectory data is used as indicators for statistical analysis. The experiment demonstrated that the cumulative probability distribution of travel time for the sample data and virtual data passed the Kolmogorov-Smirnov (K-S) test at permeabilities ranging from 10% to 100% and at significance levels of 0.01 and 0.05. Both datasets accepted the hypothesis that they originate from the same distribution. The reliability of the proposed method for generating virtual trajectories has been demonstrated.