Research On Differentially Private Location Privacy Protection In The Edge Computing

As the key foundation for LBS applications,the indoor localization has become one of the indispensable important services in our daily life.However,with the explosive growth of the number of network smart devices,if all data is transmitted to the cloud for computation and analysis will cause huge network load and computational overhead.The real-time and highaccuracy demands of the indoor localization face important challenges.How to provide highquality,strong real-time and lightweight indoor localization services while ensuring user's privacy has become a research hotspot.Therefore,in this thesis,our research focuses on the privacy problem of indoor localization in different edge computing application environments and the security of localization model.The main contributions are summarized as follows:(1)Aiming at the privacy problem of single-signal indoor localization,an indoor localization privacy protection method based on differential privacy for the pervasive edge computing environment is proposed.In this method,the complete Wi-Fi fingerprint data is first separated and isolated,preserved for privacy,and aggregated before being uploaded to the cloud.Then,the trusted localization model training is performed in the cloud platform.Experimental results show that,our method can provide ?-differential privacy while controlling the accuracy loss of the localization model to 8.9%.(2)Aiming at the privacy problem of multi-signal fusion indoor localization,an adaptive fingerprint-fusion indoor localization privacy protection method under edge computing is proposed.The method is based on a multi-level edge network privacypreserving architecture.By allocating multi-level privacy budget ?,the proposed method extends the differential privacy to the fingerprint-fusion semi-supervised extreme learning machine for indoor localization.Theoretical and comprehensive experimental results demonstrate that the proposed method can control the accuracy loss of the localization model to 2.2% while providing ?-differential privacy protection,and the extra time overhead can be ignored.(3)Aiming at the issues of compliance,legality and data islands of indoor localization under edge computing,a privacy-preserving method for indoor localization based on federated learning is presented,which conducts trusted federated training of indoor localization models in the edge computing environment.During the training process,users do not share training data,but only share model parameters to the cloud for the distributed training of localization models.With the collaboration between devices and the cloud,the localization model is iteratively learned and deeply optimized to achieve privacy protection and common benefits of multiple users.Comparing with the traditional centralized model training method and the model training method based on federated learning,the proposed method can not only provide provable ?-differential privacy but also guarantee almost lossless location accuracy and less time consumption.