Design And Optimization Of Security Mechanism For Blockchain-Based Internet Of Things System

With the progress of science and technology,the rapid development of Internet of Things(IoT)technology and application,followed by the rapid increase in the number of terminals,the amount of data in the IoT explosive growth,how to better ensure the security of data has gradually become an important issue.The federated learning algorithm,as a multiterminal cooperative machine learning algorithm,can solve the problem of data isolation while protecting data privacy.The decentralized,immutable and traceable nature of blockchain technology makes it a viable solution to the security problems of the IoT.However,the high on-chain latency and low throughput of the conventional blockchain system make it unable to be directly applied to the IoT system.The federal learning algorithm lacks the methods to detect and prevent malicious attacks.Therefore,to meet the needs of data security and privacy,we combined blockchain technology and federated learning algorithm to design a trusted federated learning system in the IoT,and optimized its performance.The specific research content of this thesis is as follows.First of all,based on the IoT the federal study node of malicious behavior,this thesis designed and implemented the Blockchain-based Trusted Federated Learning System(BCTFL).In this thesis,we design a trusted federated learning mechanism with credibility evaluation algorithm,and implement federated learning through smart contracts.The credibility evaluation method and the corresponding smart contract are designed to realize the quantitative evaluation of the credibility of nodes in the federated learning process.Then,this thesis designs a mutual credibility evaluation mechanism between leader and followers.This mechanism adjusts the credibility by evaluating the behavior of nodes in real time during the running of federated learning,and identifies and excludes malicious nodes according to the credibility,so as to ensure that the performance of the global model is not affected.In order to verify the system performance,two kinds of federated learning tasks are designed based on CIFAR-10 and MNIST datasets.The experimental results show that the system can guarantee the credibility of the federated learning process when the number of malicious nodes does not exceed the maximum of the fault-tolerant node number in the blockchain consensus protocol.Besides,the system is effective in two kinds of federal learning tasks and the system has high throughput and low system latency.In addition,on the basis of the research of trusted federated learning system,this thesis proposes a node selection algorithm based on reinforcement learning to solve the problems of system latency and model performance in federated learning of the IoT.In this thesis,the above performance problems in BCTFL are analyzed,modeled and translated into Markov decision process,and the Deep Q network algorithm is used to solve the optimal node selection problem.Based on the credibility,signal-to-noise ratio and computing resource of nodes,the algorithm dynamically selects nodes that participate in model training and aggregation,so as to make full use of the resources of devices,reduce system latency and improve the global model.Then,the performance of the proposed algorithm is compared with three benchmark algorithms.The results show that the proposed algorithm can ensure that the model converges to a lower loss function at a faster speed and obtain a lower system latency.Finally,we simulate the performance of the proposed algorithm under different kinds of federated learning tasks.The comparison results show that the performance of proposed algorithm exceeds that of the benchmark algorithm under different kinds of federated learning tasks.