摘要
准确和实时的轨迹数据发布能够为用户提供最新的交通和路况信息,有助于用户合理规划出行时间和路线,但是,位置信息的不当发布和反向推理容易泄露用户个人信息,甚至危及用户的生命安全。采用差分隐私方法添加的噪声,会导致隐私保护在数据发布和有效性方面引入不准确性。为了提高发布数据的准确性和可用性,提出了一种基于深度学习和差分隐私模型的数据发布方法,确保时空轨迹数据的安全发布。首先,设计了一种自顶向下递归划分区域的方法,并根据递归深度的增加,多维度定义隐私预算分配规则;其次,通过时空图卷积网络(T-GCN)提取数据的时间和空间特征预测隐私预算矩阵,并对区域添加Laplace噪声,实现轨迹数据的隐私保护。实验结果表明,在满足ε-差分隐私的前提下,该方法能更合理地实现轨迹的隐私保护。
Accurate and real-time trajectory data release can provide users with the latest traffic data information,and help users reasonably plan travel time and route.However,location information with improper releasion and reverse inference can easily disclose users’personal information and even endanger users’life safety.The noise added by differential privacy approach may lead to privacy protections introducing inaccuracies in data publication and validity.In order to improve the accuracy and usability of published data,a data publishing method based on deep learning and differential privacy model is proposed to ensure the safe release of spatiotemporal trajectory data.Firstly,a method of top-down recursive division of regions is designed,meanwhile the privacy budget allocation rules are defined in multiple dimensions according to the increase of recursion depth.Secondly,the temporal and spatial characteristics of the data are extracted by the spatiotemporal graph convolutional network(T-GCN)to predict the privacy budget matrix,and Laplace noise is added to the region to realize the privacy protection of trajectory data.Experimental results show that under the premise of satisfyingε-differential privacy,this method can realize the privacy protection of trajectories more reasonably.
作者
熊婧
杜鹏懿
冯晓荣
XIONG Jing;DU Pengyi;FENG Xiaorong(CEPREI,Guangzhou 511370,China)
出处
《电子产品可靠性与环境试验》
2024年第2期76-81,共6页
Electronic Product Reliability and Environmental Testing
