位置数据作为大数据中最为基础且广泛存在的数据类型，是构成众多大数据共享型应用的关键核心组件，保障用户位置的隐私安全是其开放和共享的前提。差分隐私在位置大数据隐私保护方面有其独特优势，但在保护具有多维相关特性的位置数据时面临隐私强度低于设定值的问题，成为制约差分隐私理论实用化的重要理论瓶颈。本项目研究相关性约束条件下的位置差分隐私保护理论和方法，分析位置时间、空间和属性的相互约束关系，提出位置相关性的表示和量化模型；基于最优估计理论分析现有位置差分隐私保护方法存在缺陷的本质原因，提出相关性约束条件下的位置差分隐私保护模型；探索满足特定相关系数矩阵的多维广义拉普拉斯噪声生成机理，构建高效可实施的多维噪声生成方法。本项目可以为位置隐私保护提供新的研究思路和技术支撑，将差分隐私理论从一维推广到多维相关性数据的保护方向，对差分隐私理论的深化和完善具有重要意义。

As the most fundamental and widely existing data type in big data, location data is a key component of substantial big data sharing applications. Nonetheless, preserving individual location privacy is the premise of its openness and sharing. Differential privacy has its unique advantages in privacy preserving of location big data. However, when protecting location data with multi-dimensional correlation characteristic, the privacy degree of differential privacy is lower than its setting value. It has become an important theoretical bottleneck restricting the implementation of differential privacy. This project studies differentially private theory and methods to protect location privacy under the condition of correlation constraint. It analyzes the mutual constraint relationship among location time, space and attributes, and proposes a quantitative representation and estimation model for location correlation. Based on optimal estimation theory, it analyzes the essential reason of the degradation of current privacy preserving methods under the correlation constraint. This project explores the generalized Laplacian noise generation principle to meet the specific correlation coefficient matrix, to construct an efficient and implementable multi-dimensional noise generation scheme. This project can provide new ideas and technical supports for location privacy preserving. It can not only solve the problem of location privacy preserving, but also extends differential privacy from single-dimensional correlation data protection to multi-dimensional data case, which has an important significance for deepening and improving the theory of differential privacy.