Aiming to address the lack of punishment mechanisms for malicious nodes, high communication overhead, and insufficient security in master node selection within practical Byzantine fault-tolerant consensus algorithms, this work proposes a PBFT consensus optimization algorithm based on trust and standard deviation clustering, TSD-PBFT, which aims to improve the consensus efficiency and security. Firstly, the algorithm establishes a dynamic and static trust assessment model to evaluate node behavior in real-time by monitoring node votes and participation. This process eliminates malicious nodes to enhance overall consensus efficiency and reliability. Additionally, the scores of nodes with high trust values are periodically reset to prevent individual nodes or small groups from dominating the consensus process for extended periods. Secondly, the algorithm introduces a clustering approach based on trust and standard deviation. It selects nodes with high density and strong trust as clustering centers by incorporating standard deviation, which avoids local optimal solutions. The improved K-medoids clustering algorithm groups the nodes into two layers, facilitating layered consensus and reducing communication overhead during the consensus process. Finally, the algorithm optimizes the master node selection method. Nodes at the clustering centers vote to choose the master node, giving higher voting weights to those with high trust and low standard deviation. This approach reduces the likelihood of malicious nodes becoming the master node and enhances the security and fairness of the selection process. Experimental simulation results demonstrate that, under the same network settings and number of nodes, the TSD-PBFT algorithm improves average throughput by 72.1% and reduces average delay by 50.2% compared to PBFT. TSD-PBFT also shows significant performance advantages over similar PBFT improvements, making it more suitable for large-scale consortium blockchain scenarios.