Research On Federated Spatial Data Analysis Combining Differential Privacy And Secure Aggregation

In the field of spatial data analysis,it is common to collect distributed data from millions of individuals.However,such data is often sensitive and may expose personal information（e.g.,location,trajectory,etc.）during analysis.Besides,competitions among data owners and spatial data heterogeneity may lead to spatial data islands,which significantly hinder collaborative progress.As a promise technique,federated analysis can handle the above issues,which mainly focuses on spatial data privacy,utility of result,and communication costs.This thesis researches on the privacy and security of spatial data releasing and analysis.However,data island and spatial indexing directly constrain the utility of the current methods.To address the current problems,we propose an efficient method,called HTR-Publish,to publish spatial data,and response spatial range query.In HTR-Publish,we first employ a hierarchical tree to index the reported spatial data of users in each federated iteration.And then,we use distributed differential privacy and secure aggregation to ensure privacy and security of aggregated results in each iteration.Furthermore,we rely on noise variance to adjust the merge threshold in clipping the hierarchical tree.Besides,to address the low efficiency of federated spatial data queries and the sparsity of participant encoding,a hierarchical tree-based federated spatial range query method is proposed.This method uses secure aggregation based on fault-tolerant learning to reduce communication costs and compresses encoding using the CMS structure to reduce sparsity.The experimental results of the proposed spatial data publication method show that the mean square error is below 10^-9.Our method conducts range queries on different spatial data sets,compares with different privacy budget allocation strategies and pruning threshold strategies,and achieves a relative error of about 10^-2 by comparing the relative error of the results.The experimental results demonstrate the robustness of the proposed solution.