With the expansion of the Industrial Internet of Things (IIoT), the dynamic characteristics of IIoT device nodes and the diversity of available resources present significant challenges to blockchain consensus mechanisms. Existing Practical Byzantine Fault Tolerance (PBFT) algorithms struggle to adapt to the inherent dynamic node changes in IIoT applications. Therefore, this paper proposes an improved consensus algorithm based on a comprehensive reputation model and dynamic clustering, denoted as ICD-PBFT (Improved Consensus based on Dynamic Reputation and Incremental Clustering for PBFT). The algorithm enhances performance through a threefold optimization mechanism: First, it incorporates a Multi-Dimensional Reputation Model (MD-RM) to optimize the hierarchical filtering of node reputation values. Second, it utilizes a combination of geographic distance and resource utilization to perform dynamic hybrid clustering and employs a Resource Adaptation Dynamic Clustering Strategy (RA-K-means) to optimize sub-cluster partitioning and reduce cross-domain communication latency. Third, it employs a hierarchical consensus architecture that decomposes global consensus into intra-cluster pre-consensus and inter-cluster confirmation, thus streamlining the consensus process. Simulation results show that the ICD-PBFT consensus algorithm outperforms existing PBFT algorithms in terms of consensus latency, throughput and consensus success rate.