Research On PBFT Consensus Algorithm Based On Anomaly Detection And Reputation Model

As a distributed storage technology,blockchain has the characteristics of decentralization,tamper proof and traceability,which can effectively solve the trust problem.Blockchains can be applied not only in digital currency,but also in finance,medical,logistics and cross-border payment.Among them,the consensus node in the consortium blockchain needs to pass the authentication,so its network environment is relatively stable and reliable.Nowadays,many commercial applications prefer to use consortium blockchains,but the existing consensus algorithms of consortium blockchains still have abnormal and security problems,so how to make the PBFT consensus algorithm execute safely and efficiently is an urgent problem to be solved.PBFT is a common consensus mechanism in consortium blockchain.Although PBFT has certain Byzantine fault tolerance,there are still two problems:1.In the process of consensus,the existence of malicious nodes/behaviors will still affect the efficiency of consensus in the consortium blockchains,and then affect the TPS of the consortium blockchains network.For the node anomaly problem in blockchains network,the research has focused on using machine learning to detect malicious behavior in public blockchains,and it is not common in the context of consortium blockchains.2.PBFT algorithm may choose malicious nodes as the primary nodes when selecting the primary nodes.This makes the primary node exposed in advance,increases the possibility of the primary node being attacked,and reduces the security of PBFT algorithm.In order to select safe and efficient nodes as the primary node of PBFT consensus mechanism,the existing research introduces credit mechanism,but some of the existing credit mechanisms have some increasing reputation value,which leads to the problem of centralization.In order to solve the above two problems,this paper mainly does the following works:1.In order to improve the reliability and consensus efficiency of consortium blockchains,this paper proposes KNN,naive Bayes,SVM and CNN algorithm to detect the anomaly of PBFT consensus algorithm,judge whether there is malicious delay,and then identify malicious delay nodes.2.In order to avoid the centralization of node’s reputation in reputation model,this paper uses the results of anomaly detection in reputation model to evaluate the reputation value of nodes more accurately.By introducing random number into reputation model,the selection of primary node in PBFT has randomness.Through the experimental analysis,the anomaly detection model can detect whether there is malicious delay in the consensus process with high accuracy,and can accurately identify the nodes with malicious delay.The random number reference makes the probability that the node is selected as the primary node increases with the increase of reputation value,which effectively avoids the centralization problem caused by selecting the node with the largest reputation value as the primary node.