Research On Privacy Protection Of Internet Of Things Data Publishing Based On Differential Privacy

In recent years,Internet of things has been applied to more and more fields,bringing convenience to our life.In order to excavate the value contained in the Internet of things system,we need to publish the data for external use and research.With the development of intelligent devices,the nodes that collect information in the Internet of things system expand from traditional hardware devices to entity nodes represented by intelligent devices and their users,and the information collected becomes more complex and private.Therefore,before the Internet of things data is released,it is necessary to protect its privacy.In order to solve the privacy protection problem of dynamic set-valued data publishing in the Internet of things system,this thesis uses the differential privacy protection model to process the privacy information in the system,and proposes a privacy protection algorithm for continuously publishing dynamic data in the Internet of things.First of all,in order to deal with the challenge of dynamic,a dynamic data set preprocessing algorithm is proposed.Secondly,it optimizes and improves the existing algorithm from the aspects of algorithm performance and effectiveness,adopts the idea of classification protection,constructs the initial classification model by using the concept of classification tree,and proposes the classification tree modeling algorithm.Finally,a dynamic update algorithm is proposed,which adds the dynamic data set records to the leaf nodes of the classification tree,and publishes the count results with noise.In order to improve the performance of the algorithm and the effectiveness of the results,this thesis optimizes each step of the algorithm.Firstly,this thesis uses the idea of time window and stratified sampling to solve the problem of dynamic data set.It publishes the data set with time window as a unit,and screens the similar data sets in the same time window,so as to improve the processing efficiency of the algorithm.Secondly,the thesis uses B+ tree as the classification tree mode,which can not only avoid the waste of privacy budget caused by empty nodes,but also sort according to the number of items,and lay a good foundation for the subsequent allocation of privacy budget.In order to ensure the privacy of the classification tree structure,the index mechanism is used to select the optimal order to construct the B+ tree.The concept of privacy budget node is introduced,and adaptive noise adding method is adopted to control the cost of privacy budget and ensure the balance between data availability and privacy.Finally,the existing classification tree is updated to avoid repeated classification.Setting a threshold to determine whether to reallocate the privacy budget node can not only control the budget overhead,but also avoid the data error caused by noise accumulation.In the process of publishing,this thesis uses the characteristics of B+ tree leaf nodes connected in pairs to output the published results at one time,which further improves the efficiency of the algorithm.This thesis analyzes the rationality of privacy budget allocation from the perspective of theory,and analyzes the time complexity of the algorithm.Based on three real datasets,different parameters are set to test the proposed algorithm.Through the analysis of experimental results and comparison with other algorithms,the performance and effectiveness of this algorithm are verified.