Aiming at the problems of unstable primary nodes and high latency in the existing Byzantine fault-tolerant consensus algorithms based on grouping strategies, this paper proposed a PBFT consensus algorithm based on communication delay clustering and node reputation(CD-PBFT). Firstly, the algorithm designed a new clustering algorithm based on communication delay to group the nodes in the network, fused the communication delay into the Euclidean distance formula, and allowed the nodes in the system to be clustered based on the mixture of distances, which ultimately minimized the sum of the delays in each cluster, reduced the communication overhead, and improved the consensus efficiency. Secondly, the algorithm proposed a reputation model based on comprehensive evaluation, which took into account the node delay index, consensus behavior and historical reputation for evaluate the reputation of nodes, and rewarded and punished nodes based on their behavior and delay differences. Finally, it optimized the primary node selection method and established a primary node selection mechanism based on node stability and reputation model. After obtaining the reputation value of the node through the reputation model, it introduced the variance to measure the fluctuation of the node's reputation, and selected the node with high reputation and small variance to be the primary node, which improved the security of the primary node. Experimental results show that compared with PBFT, the algorithm improves the average throughput by 126.8% and reduces the average delay by 68.3%. Meanwhile, compared with the existing clustering-based PBFT algorithm, CD-PBFT has more obvious performance advantages and can be more flexibly applied in large-scale node federation chain scenarios.