Mobile devices furnish users with various services while on the move, but also raise public concerns about trajectory privacy. Unfortunately, traditional privacy protection methods, such as anonymity and generalization, are not secure because they cannot resist attackers with background knowledge. The emergence of differential privacy provides an effective solution to this problem. Still, the existing schemes are almost designed based on the collected aggregate historical data (so-called static trajectory privacy protection), which are not suitable for real-time dynamic trajectory privacy protection of mobile users. Furthermore, due to the complexity and redundancy features of the full trajectory data, the efficiency and accuracy of the privacy protection model are significantly limited by the existing schemes. In this paper, we propose a new differential privacy scheme base on the Recurrent Neural Network for Dynamic trajectory privacy Protection (RNN-DP). We firstly introduce a recurrent neural network model to handle the real-time data effectively instead of the full data. Secondly, we novelty leverage the dynamic velocity attribute to form a quaternion to indicate the status of the users. Moreover, we design a prejudgment mechanism to increase the availability of differential privacy technology. Compared with the current state-of-the-art mechanisms, the experimental results demonstrate that RNN-DP displays excellent performance in privacy protection and data availability for dynamic trajectory data.

移动设备在旅途中为用户提供各种服务，但也引起公众对轨迹隐私的关注。不幸的是，传统的隐私保护方法（例如匿名性和泛化性）并不安全，因为它们无法抵御具有背景知识的攻击者。差异隐私的出现为该问题提供了有效的解决方案。尽管如此，现有方案几乎都是基于收集的汇总历史数据（所谓的静态轨迹隐私保护）来设计的，这不适合移动用户的实时动态轨迹隐私保护。此外，由于整个轨迹数据的复杂性和冗余性，现有方案严重限制了隐私保护模型的效率和准确性。在本文中，我们基于递归神经网络的动态轨迹隐私保护（RNN-DP），提出了一种新的差分隐私方案。我们首先介绍一个递归神经网络模型，以有效地处理实时数据而不是完整数据。其次，我们利用动态速度属性来形成四元数以指示用户的状态。此外，我们设计了一种预判机制来提高差异隐私技术的可用性。与当前的最新机制相比，实验结果表明RNN-DP在动态轨迹数据的隐私保护和数据可用性方面显示出出色的性能。我们首先介绍一个递归神经网络模型，以有效地处理实时数据而不是完整数据。其次，我们利用动态速度属性来形成四元数来指示用户的状态。此外，我们设计了一种预判机制来提高差异隐私技术的可用性。与当前的最新机制相比，实验结果表明RNN-DP在动态轨迹数据的隐私保护和数据可用性方面显示出出色的性能。我们首先介绍一个递归神经网络模型，以有效地处理实时数据而不是完整数据。其次，我们利用动态速度属性来形成四元数来指示用户的状态。此外，我们设计了一种预判机制来提高差异隐私技术的可用性。与当前的最新机制相比，实验结果表明RNN-DP在动态轨迹数据的隐私保护和数据可用性方面显示出出色的性能。