基于密度的噪声应用空间聚类算法的差分隐私轨迹保护机制

目前大多数的轨迹隐私保护方法对轨迹的形状相似性考虑并不充分，并且容易忽略各轨迹点之间的时序相关性，导致生成的干扰轨迹可用性不高。为了解决这些问题，提出了一种基于密度聚类算法(density based spatial clustering of application with noise, DBSCAN)的差分隐私轨迹保护机制。首先，使用DBSCAN算法对数据进行聚类分析，降低数据集中噪声点对聚类效果的影响；其次，根据用户活动轨迹点的时序关系，生成位置转移概率矩阵，利用差分隐私的方法确保生成的干扰轨迹点与真实轨迹点具有相似的位置转移概率；最后综合考虑差分隐私预算和弗朗明歇距离(Fréchet distance)对轨迹相似性的影响，选取位置干扰点。通过仿真实验分析，本文的方案在效率上具有明显的优势，并且生成的干扰轨迹与真实的位置轨迹相比具有较高的形状相似性。

Differential Privacy Trajectory Protection Mechanism Based on DBSCAN Clustering Algorithm

Most of the current trajectory privacy protection methods does not consider the shape similarity of trajectory sufficiently, and it is easy to ignore the timing correlation of each point on the trajectory, which results in the low availability of the generated interference tracks. To solve these problems, a differential privacy trajectory protection mechanism based on density clustering algorithm was proposed. Firstly, DBSCAN algorithm was used for clustering analysis of data to reduce the influence of noise points in data set on clustering effect; Secondly, the position transition probability matrix was generated according to the time sequence relationship of user''s activity locus points, and the differential privacy method was used to ensure that the generated interference locus points have similar position transfer probability to the real locus points; Finally, considering the influence of differential privacy budget and Fréchet distance on trajectory similarity to select the location interference point. The simulation results show that the proposed scheme has advantages in efficiency obviously, and the generated interference trajectory has higher shape similarity compared with the real position trajectory.