Research On The Extensibility Of Blockchain PBFT Consensus Algorithm

Consensus protocols play a vital role in blockchain,solving the problem of reaching a consensus on the outcome of a transaction among involved participants.However,with the increasing complexity of the network environment and the growing number of network users,the development of blockchain is gradually limited by the efficiency,security and reliability of consensus protocols.At present,the classic Practical Byzantine Fault Tolerant algorithm PBFT is widely used,but in its consensus process,nodes need to interact with all other nodes,that is,the communication complexity is(9)~2),when the number of nodes increases,it will lead to system communication Congestion and throughput are reduced.In addition,the malicious behavior of the master node will cause frequent view switching and increase the burden on the system.This also makes the PBFT algorithm unable to be applied to large blockchain networks.In response to the above problems,this thesis proposes a consensus algorithm CBFT based on the reputation model,and designs a dynamic sharding strategy based on deep reinforcement learning to improve the scalability and security of the blockchain system.The specific research content of this thesis is as follows:First,aiming at the high communication cost of the PBFT consensus algorithm,an improved blockchain consensus mechanism CBFT based on the reward and punishment strategy is proposed.In CBFT,the nodes have different credibility according to the consensus performance of the nodes.At the same time,the nodes are divided into consensus nodes and candidate nodes.Through the reliable node selection strategy,the probability of normal nodes joining the committee to participate in the consensus is increased,the number of malicious nodes in the committee is reduced,and the number of malicious nodes in the committee is improved.safety.At the same time,a malicious node detection mechanism based on machine learning is designed,and the node reliability is evaluated in all aspects through the historical performance and current state of the node consensus process.The experiment proves the validity of the reputation model and the reliable node selection mechanism,and can compare A good screening system for malicious nodes.Second,a blockchain dynamic sharding strategy based on deep reinforcement learning is proposed.Blockchain sharding divides the network into different consensus groups,and different consensus groups perform consensus in parallel to improve throughput.However,the static blockchain sharding strategy can easily lead to system centralization,and at the same time,groups may be taken over by malicious nodes,which cannot effectively deal with the dynamic blockchain environment,and the system parameters in the dynamic blockchain environment are also difficult to determine.Based on this,this thesis analyzes the main factors affecting the throughput of the blockchain,and establishes a corresponding model,using deep reinforcement learning to dynamically fragment the blockchain nodes,and forms a dynamic layer on the basis of dynamic fragmentation,each The consensus group elects leading nodes through the reputation value to join the main consensus layer,aiming to improve the security and throughput of the system,achieve a balance between security and throughput,and obtain the best parameters of the system at the same time.In this thesis,block size,sharding period and number of groups are used as optimization parameters,and the dynamic changes of the network are considered at the same time.Experiments prove that the sharding scheme in this thesis can effectively improve system throughput and ensure system security.