| Recently,with the rapid development of embedded technology and mobile communication technology,the number of Io T devices has been proliferating.The everything-interconnected network architecture and the development trend of computing network marginalization have brought development momentum to a new round of technological revolution and industrial change worldwide.As the “core driver”of Internet of things(Io T),the secure and efficient transmission of Io T data are the important technical guarantee to promote the construction and development of automation,intelligence,and information infrastructure.To address the security challenges regarding the single point of failure(SPOF),data security,and trust issue under the Io T architecture,the integrated data-sharing architecture of blockchain and Io T has received extensive attention from academia and industry and has broad application prospects.However,the limited computing resources and storage capacity of Io T devices,as well as the high computing and communication overhead of blockchain prevent the blockchain from being deployed in the large-scale Io T scenarios.Particularly,the design of network architecture,blockchain node computing performance,and system scalability are crucial challenges in complex application scenarios with low latency and high-reliability requirements.Meanwhile,the data services,node computing and communication resources in the Io T scenario usually have stochastic and dynamic characteristics.In that case,how to model the specific requirements for different tasks and the complex state transition of the network resources,and deploy optimization algorithms to jointly and dynamically consider system performance will become one of the most important topics of future research.In addition,the issues of user grouping,task scheduling,privacy protection,and resource allocation in task-driven data sharing scenarios(e.g.mobile crowdsensing and collaborative computing)have not been considered in depth.In order to address the above issues and challenges,we pay more attention on the research about the key technologies for Io T data sharing based on blockchain and deep reinforcement learning.The main research works and contributions are listed as follows.(1)Research on scalability and load balancing for sharded-blockchainenabled Io T data sharing system.Focus on the features of the Io T data sharing scenario in large-scale devices,dynamic and stochastic Io T tasks and small volume of communication data,in this dissertation,we propose a sharded-blockchain-enabled Io T to solve the problems of the single point of failure,data security and privacy issues in the traditional centralized system architecture.Based on sharding,the blockchain consensus nodes are divided into different consensus zones,the large number of transactions within the blockchain network can be validated in parallel within multiple shards,so as to effectively support the large-scale systems.However,improving system throughput,reducing consensus latency,and ensuring inter-shard computational load balancing under a random node assignment policy become important technical challenges in this sharding architecture.Therefore,we propose a deep reinforcement learning-based approach to optimize the performance of a sharded blockchain.Based on the modeling of the computational capacity of consensus nodes within the blockchain and the theoretical analysis of the potential computational load of the intra-shard consensus,the access between the sharded node network and the data transaction pools,as well as the dynamic adjustment of the underlying blockchain parameters are formulated as a joint optimization problem.In other words,the state space and action space of the system are formulated as the Markov decision process(MDP).Meanwhile,the system throughput,latency and computational load of the sharded blockchain are formulated as a reward function,so as to select the optimal access strategy between the sharded node network and the data transaction pool as well as the blockchain parameter adjustment strategy in each decision episode.On the other hand,a deep Q network(DQN)is trained to maximize the reward function to obtain the optimal policy.The simulation results illustrate that the proposed optimization framework can improve the throughput of the blockchain significantly compared with other existing schemes and can efficiently address the loadbalance issue in the sharded blockchain.(2)Research on hybrid-blockchain architecture design and task allocation for mobile crowdsensing data sharing.Faced to the problems of data security and privacy leakage based on inference attacks in the MCS data sharing scenario,in this dissertation,we propose a novel hybrid-blockchain-based mobile crowdsensing(MCS)platform.On the one hand,the open participation feature based on the public chain enables the recruitment of MCS applicants.On the other hand,based on the access control of the permissioned blockchain,the illegal access of malicious users can be avoided.By adopting a hybrid blockchain,we combine the advantages of both public and consortium blockchains.Meanwhile,the differential-privacy-based noise perturbation method is adopted to reduce the risk of privacy exposure while data sharing.In addition,to address the issue regarding the energy consumption problem of data aggregation in large-scale MCS scenarios,a task-offloading scheme based on the combination of cloud computing and mobile edge computing is proposed to enable the MCS platform to execute more computational tasks under a low energy consumption.Moreover,a deep reinforcement learning-based task scheduling and resource allocation optimization framework is proposed to address the resource competition problem of multiple task requestors in the same platform and the optimization of the task offloading strategy.Through the problem formulation based on MDP,the task scheduling process is optimized by a double deep Q learning approach,where the number of participants,computational costs and task benefits are all taken into consideration.Then,by jointly considering the data volume,computational complexity,privacy level and other characteristics of the task,the proposed scheme can maximize the utility of the MCS platform while minimizing the task execution latency.(3)Research on scalability and computational performance of blockchainenabled Io T data sharing architecture in collaborative computing.In terms of the potential assemblage characteristics driven by the collaborative tasks among Io T users,in this dissertation,we propose a novel user-grouping-based blockchain sharding strategy for collaborative computing Io T.By taking the selforganization characteristics of Io T users as the basis for blockchain sharding and adopting the K-means-based user grouping method,the potential association structure can be accurately detected while the proportion of cross-shard transactions(CSTs)can be reduced.Considering the high computational consumption and communication latency of the existing synchronous cross-shard consensus,we propose an asynchronous PBFT-based cross-shard consensus mechanism,which can effectively reduce the computation and communication overhead of the consensus of CST by splitting the atomic operation into two parts for consensus asynchronously.In addition,a novel approach combining deep reinforcement learning and machine learning is proposed to dynamically select the optimal clustering initialization parameters to guide K-means to obtain optimal user grouping results.Meanwhile,based on the MDP,the main parameters affecting blockchain performance are jointly optimized to maximize the system throughput by a double DQN approach under the constraint of the consensus security and the block finality latency.Simulation results demonstrate that the proposed scheme can efficiently reduce the proportion of CST and the computational overhead of cross-shard consensus while ensuring the quality of Io T user grouping,and can improve the scalability of the blockchain significantly while guaranteeing consensus security. |
