Federated blockchain systems are widely used in scenarios such as finance and logistics. However, the practical Byzantine algorithm (PBFT) applied to this blockchain system suffers from low scalability and high communication cost, which hinder the application of the blockchain system in large-scale scenarios. To address these problems, a dynamic multi-organizational practical Byzantine fault-tolerant algorithm, the K-PBFT consensus algorithm, is proposed. The improved K-Means algorithm is proposed to divide the nodes into multiple autonomous organizations based on the time delay of the nodes and the communication distance between them, and each organization communicates with each other through the organizational representative nodes. When a new node joins, the algorithm assigns it to the most reasonable organization according to its characteristics. The K-PBFT algorithm also improves the security of the system by setting a reputation mechanism to identify honest nodes from malicious nodes in the system. The node tenure mechanism is introduced so that each honest node in the blockchain has the opportunity to act as the representative node or master node of the organization. Experimental results show that the communication complexity of the K-PBFT algorithm is reduced by 75% compared with the PBFT algorithm, and when the number of nodes is 100, the latency is reduced by 210 ms and the throughput is increased by 100% compared with the PBFT algorithm. And in a high latency environment, the latency is reduced by 20% and the throughput is improved by 17% compared to the improved PBFT algorithm based on reputation grouping when the number of nodes is 100.