Differential Privacy Publication Of Streaming Data For Wearable Devices

The wearable device continuously collects human health data through its own sensors and sends it to the central data server.The server releases these data to researchers for analysis and mining of valuable information,so as to provide more scientific advice and guidance to people's lives.Differential privacy is a privacy protection technology with strict proof and adjustable privacy,which is often used in privacy protection data publication scenarios.The existing differential privacy method is applied to the streaming data release of wearable devices,which cannot meet the privacy and high-availability requirements of data publication.Considering the characteristics and application scenarios of health streaming data,and designing a differential privacy protection publication scheme for health streaming data to improve the privacy and utility of the released data.The intermittent stream data of human health increases indefinitely with the extension of time,and it has the characteristics of autocorrelation and periodic change.Aiming at this feature,this thesis develops a sliding window-based Differential Privacy Autocorrelation Intermittent stream data publishing algorithm(DPAI).The DPAI proposes period sensitivity to replace the global sensitivity in traditional differential privacy,and adds correlation noise to the intermittent stream data with autocorrelation.At the same time,a sliding window is introduced to process and publish the infinitely growing intermittent stream data.A comparative experiment was conducted on two real datasets.The experimental results show that under the same privacy budget,the DPAI algorithm can reduce the error by about 25%and provide a higher degree of privacy protection.Data collection and release under untrusted servers can easily lead to the disclosure of personal information,the existing differential privacy data publication scheme introduces distributed agents to avoid privacy leakage,but this scheme adds too much noise at each agent,reducing data availability.To solve the above problems,this thesis develops a distributed agent-based Differential Privacy Real-time Stream data publishing algorithm(DPRS).The DPRS algorithm proposes a similarity grouping mechanism.At each agent,further combination is performed by calculating the Pearson correlation coefficient between the groups.The mechanism adds noise to the statistical data in the new group and publish it,which reduces the noise added to the published data to a greater extent.Experimental results show that under the same privacy budget,sliding window size,and the same number of agents,the DPRS algorithm can significantly reduce the error of release data.Finally,based on the interactive differential privacy protection framework,the proposed algorithm is applied to actual data publication.This paper designs and implements a prototype system of differential privacy publication for wearable devices' streaming data.The system can release the real human health stream data after being disturbed,and show the release results to data analysts in a variety of ways.And data analysts can analyze data characteristics from the published results and obtain valuable information.