Due to its unique advantages, this new field of blockchain has developed rapidly in the fields of communication, artificial intelligence, and the Internet of things in recent years. The consensus algorithm is a key part of the blockchain field, and the practical Byzantine Fault Tolerant Algorithm (PBFT) Algorithm This type of algorithm is better in terms of security and fairness because it can resist Byzantine problems, but the performance of this algorithm is poor when performing consensus among large-scale nodes, and the communication overhead will increase sharply with the increase of nodes, reduced reliability, and excessive time delay. For these problems, this paper proposes a dynamic consensus algorithm (RF-PBFT) based on a reputation mechanism to divide clusters. First of all, according to the performance of the nodes, the Fast Affinity Propagation (Fast AP) algorithm is used for sub-cluster consensus, which greatly reduces the complexity of communication. Secondly, a dynamic adjustment mechanism based on node reputation is proposed to reduce the influence of Byzantine nodes doing evil and increase the reliability and security of the algorithm. Third, based on this algorithm model, the consensus algorithm process and view changes are also optimized to reduce delays and improve consensus efficiency. Experimental results show that RF-PBFT can significantly improve system performance and improve consensus efficiency.