| With the rapid rise of Internet of Things(IoT)applications and the number of IoT devices,how to guarantee the data security of IoT systems has become a key problem.With the immutability and decentralized characteristics,blockchain technology can effectively address the data security raised in IoT applications.However,due to the IoT features,such as dynamic environment,huge data,and restricted device performance,applying blockchain directly to the IoT system will result in various issues,including increased chain delay and lower throughput.The high requirements of blockchain for computing,storage,and communication resources contradict with the limited capabilities of IoT terminals.Therefore,to guarantee the data security and performance of IoT applications,this thesis investigates the block production and consensus in blockchain systems to reduce uploading latency of IoT data and increase transaction throughput.The following are the specific research contents.An intelligent transaction migration technique based on the Deep Deterministic Policy Gradient(DDPG)algorithm is proposed for the RAFT consensus of private blockchain in IoT systems.This method could balance the burden of clusters by blockchain transactions among base station nodes,and relieve the problem of imbalanced data load and node performance among IoT clusters.The optimization training is carried out with the main objective of reducing the total uplink delay,and finally the optimal transaction migration strategy is determined by using the DDPG algorithm in reinforcement learning.The results show that the proposed smart transaction migration scheme could reduce the on-chain delay by 29%and 58%,respectively,compared with the random transaction migration strategy and the static strategy.Taking into account the high computation and communication requirements of the RAFT consensus mechanism,this thesis explores the performance potential of IoT nodes and proposes a dynamic weighted RAFT(WRAFT)consensus mechanism that is more suitable for IoT applications.In order to relieve the problem that each node in the RAFT consensus can only be randomly elected as the leader node with equal probability,the weight is dynamically calculated according to the performance of each node by leveraging the reinforcement learning DQN algorithm,so that the node with better performance has a greater probability of being elected.The leader node thus undertakes heavier computing and communication tasks.The simulation results show that the proposed WRAFT consensus scheme can reduce the data forwarding delay by 27%compared with the original RAFT consensus mechanism. |
