目前的区块链系统在吞吐量、交易处理时间、共识算法的安全性等指标方面还不是很完美，因此区块链系统还暂时无法完全取代传统互联网提供的各类功能。共识算法是区块链应用中使用的一种协议，用于确保参与共识的所有节点都能就特定交易达成一致。因此研究具有容错能力，可以抵御恶意节点对共识影响的高效、安全共识算法，对于区块链的发展和未来广泛应用具有极其重要的意义。针对上述问题，本文做了如下工作：

    （1）提出了一种基于节点权限的两阶段领导人选举方案，解决了 Raft 算法领导人选举耗时受网络分区、节点数量过多影响的问题。改进方案为选举新增预投票过程，根据节点的表现，为集群节点新增信任属性，并根据信任属性计算节点的选举权限。在领导人选举过程中，通过比较节点权限，解决由于节点数过多导致无法通过一轮选举选出领导人节点的问题。通过为选举增加预投票过程的方式，在正式选举开始前判断发起选举节点是否是被网络分区孤立的节点。解决共识过程可能存在被网络分区等非理想状态网络环境干扰的问题。此外，通过对方案活性以及拜占庭容错性的分析，证明该方案在非理想状态的网络环境下依然可以正常工作。最后，通过仿真实验对方案的效率、性能进行全面验证。

    （2）提出了一种基于验证委员会的日志复制方案，解决了 Raft 算法无法抵御恶意节点对共识结果的影响、干扰问题。考虑到系统可能存在恶意节点，算法根据信任值和节点活跃程度选出一定数量的节点组成验证委员会。系统发送的信息需要经过验证委员会的验证才可以生效，解决共识结果遭网络中恶意节点影响、干扰的问题。此外，通过概率推导对方案安全性进行分析，证明了本课题所设计的日志复制方案在具有拜占庭节点的系统中是安全可用的。最后，通过仿真实验证明，相较于原方案来说，改进方案在允许系统存在拜占庭节点的同时，吞吐量性能更优，用户感知延迟更低，单位时间内处理的客户端请求数更多。

The current blockchain system is not yet perfect in terms of throughput, transaction processing time, security of consensus algorithms, and other indicators. Therefore, blockchain systems are currently unable to completely replace various functions provided by traditional internet. So the blockchain systems are temporarily unable to completely replace various functions provided by the traditional Internet. The consensus algorithm is a protocol used in blockchain applications to ensure that all nodes participating in the consensus can reach an agreement on the certain transaction. Therefore, researching efficient and secure consensus algorithm with fault tolerance capabilities that can resist the impact of malicious nodes on consensus is of great significance for the development and future wide application of blockchain. In response to the above issues, this thesis has done the following work:

(1)A two-stage leader election based on node permissions is proposed to solve the problem of Raft algorithm leader election time being affected by network partitioning and excessive number of nodes. The improved algorithm adds a pre-vote process for elections, adds trust values for cluster nodes based on the performance of the node, and calculates the election permissions of the nodes based on the trust values. During the leader election process, the problem of being unable to elect the leader through one round of leader election in a cluster with too many nodes can be solved by comparing node permissions. By adding a prevote process for the election, it can be determined whether the node initiating the election is isolated by the network partition before the formal election begins. Thereby solving the problem that the consensus process may be disturbed by non-ideal network environments such as network partition. In addition, through the analysis of the algorithm activity and Byzantine fault-tolerance, it is proved that the algorithm can still work normally in a non-ideal network environment. Finally, the efficiency and performance of the algorithm are comprehensively verified through the experiment.

(2)The log replication algorithm based on a verification committee is proposed to solve the problem that Raft algorithm cannot resist the impact and interference of malicious nodes on consensus results. Considering that there may be malicious nodes in the system, the algorithm selects a certain number of nodes to form a verification committee based on trust values and node activity. The transaction sent by the system needs to be verified by the verification committee before it can take effect, which can solve the problem of consensus results being affected or interfered by malicious nodes in the network. In addition, by analyzing the security of the algorithm through probabilistic knowledge, it can be proved that the log replication algorithm designed in this thesis is safe and available in systems with Byzantine nodes. Finally, the experiment have shown that compared to other algorithm, the improved algorithm has better transaction throughput, lower user-perceived latency, and more transactions per unit time while allowing the system to have Byzantine nodes.