Abstract: Byzantine faulttolerant consensus algorithms in blockchain face significant challenges with the increase in the number of nodes, including arbitrary leader election and high communication overhead. Existing solutions predominantly  focus on optimizing a single dimension of node grouping, but in complex network environments, how to integrate multidimensional factors for effective node grouping and when to trigger regrouping remain unresolved. To address these issues, this paper proposes a dynamic networkaware driven multidimensional adaptive grouping HotStuff (DMAGHotStuff) algorithm. Firstly, a dynamic reputation regulation mechanism is designed, incorporating reward and punishment strategies. This mechanism evaluates node reputation based on historical behaviors and communication characteristics, while considering factors such as geographic location and network latency, effectively reducing misjudgments caused by network issues and minimizing the impact of Byzantine nodes. Secondly, a multidimensional Kmeans grouping consensus algorithm is proposed, classifying nodes based on diversereputation levels and communication features, which significantly enhances the system’s tolerance to Byzantine nodes. Finally, a momentum monitoring system is used to monitor network status changes, isolating potential Byzantine nodes and determining the optimal timing for regrouping, thereby achieving more accurate adaptive grouping decisions. Experimental results demonstrate that DMAGHotStuff outperforms HotStuff, PBHS, and other HotStuff variants with different grouping strategies in terms of both latency and throughput.

Key words: blockchain, consensus algorithm, HotStuff, grouping, selfadaptation

摘要： 区块链拜占庭容错共识算法在节点数量增加时，存在领导节点选举随意、通信开销大等问题.为了解决这些问题，现有方案多聚焦于单一维度的分组优化.但在复杂网络环境中，如何综合多维度因素对节点进行合理分组，以及何时需要重新分组仍未得到充分解决.针对上述问题，提出了一种动态网络感知驱动的多维自适应分组HotStuff(DMAGHotStuff)算法.首先，设计一种信誉动态调控的奖惩机制，通过历史行为和节点通信特征值确定节点信誉度，评分时考虑节点的地理位置和时延因素，有效减少因网络问题而导致的误判并减少了拜占庭节点的影响.其次，考虑节点的通信特征值，设计多维度的Kmeans分组共识算法，对不同信誉的节点进行分类，显著增强了系统对拜占庭节点的容错能力.最后，采用动量监控系统的变化，在网络状态变化时隔离潜在的拜占庭节点，确定重新分组的时机，实现了更精准的自适应分组决策.实验结果表明，DMAGHotStuff算法在延迟和吞吐量方面优于HotStuff,PBHS及其他基于不同分组维度的HotStuff算法.

关键词: 区块链, 共识算法, HotStuff, 分组, 自适应