Research On Secure And Reliable Data Sharing Strategy In Internet Of Things

As an important part of the new generation of information technology,the Internet of things(Io T)is an information carrier based on the Internet,which make all independent entity objects form an interconnected network.Internet of things technology is widely used in many fields such as industry,agriculture,transportation,medical treatment,tourism and so on.In order to improve the service quality of Io T applications,researchers have carried out in-depth research.Among them,data analysis is a reliable and effective method.Especially with the vigorous development of artificial intelligence,the value of data is becoming more and more important.Data analysis requires a large amount of relevant data and powerful computing resources,which is difficult for entities with limited computing and storage capacity,especially in the case of scattered data.Therefore,these scattered data need to be aggregated through data sharing.For example,in the medical field,the clinical data of patients owned by each medical unit is usually limited,which affects the treatment effect to a certain extent.By sharing the data of each medical unit and aggregating and analyzing the health files,including treatment information,physical examination information and other valuable data,we can provide more targeted treatment plans for patients,Improve the treatment effect.However,data sharing in Io T faces the following problems.Firstly,due to the estrangement between interactive entities and the sensitivity of shared resources,it is difficult to establish mutual trust between entities,so the reliability of shared data can not be effectively guaranteed;Secondly,the Io T is open,and there is a huge threat of data privacy disclosure in the process of data sharing,which leads to the low enthusiasm of data owners for data sharing.In order to solve the above problems,this paper focuses on the problems of privacy disclosure and data unreliability in data sharing.The main research work is summarized as follows:(1)Aiming at the problems of privacy disclosure and unreliable data in data sharing,we propose a data sharing method based on federated learning and deep reinforcement learning.This scheme transforms the traditional shared source data into shared model parameters,which protects the privacy and security of data providers.Then,we use the asynchronous federated learning algorithm according to the subtask,and use deep reinforcement learning to select the participants with reliable data and strong computing power in each subtask,so as to ensure the reliability of shared model parameters.Because the final data request task is not visible to all participants,the task privacy of the data requestor is also protected.Finally,considering the characteristics of homomorphic encryption that allows direct operation of ciphertext,we introduce homomorphic encryption algorithm in the aggregation process of model parameters of federated learning to further protect the privacy and security of data providers.However,considering the cost of encryption,this method also increases a certain time complexity.Simulation experiments verify the effectiveness of the proposed scheme,and the selected participants have higher accuracy of the training model results.(2)Privacy reasoning in data sharing mechanism for model parameter migration,and in order to ensure the reliability of shared data,we propose an Internet of things data sharing method based on mutual supervised federated learning and blockchain,which improves the reliability of the data provided by the participants by mutually verifying the model gradient provided by other participants in each round of global model training,and eliminating the nodes with the lowest accuracy or the largest error in each round.But while this method brings benefits,it also increases the cost of time.In order to further protect the privacy and security of data nodes,we introduce differential privacy into each round of model training,and each participant executes a gradient descent algorithm with differential privacy.Finally,we store the index of each round of verification results of federated learning on the blockchain to prevent denial or tampering.In order to encourage the participation of federated learning,we design a reward mechanism,taking the verification results as the standard to measure the reward of data nodes.The simulation results show that the proposed scheme is excellent in effectiveness and security.(3)Considering that the implementation of gradient descent with differential privacy will inevitably affect the training efficiency of the model,and in order to increase the verifiable range on the basis of(2)to ensure the reliability of shared data,we propose an data sharing method based on verifiable federated learning.Different from(2),in this method,the data team completes the data sharing task,that is,the data team completes the training of the global model(the global model is the model finally delivered by the data team to the data requester).The data team is composed of various participants with data,and there will be team leaders within the data team.The data requester will decide the reward and punishment for the data team according to the received global model performance,such as accuracy or error.In addition,the "mortgage punishment" mechanism is introduced into the data team to record each training process in data sharing on the blockchain,so as to further punish the members who provide unreliable data.That is,the data team can further manage and supervise the members,so as to ensure that the team members can complete data sharing reliably.Finally,in view of the trust relationship within the data team,we introduce the differential privacy technology into the global model instead of every round of model update.By adding disturbance to the global data model,we can prevent dishonest data requesters from inferring the privacy of data team members from the obtained model and further protect their data privacy.Simulation results show that this method has high accuracy and feasibility.