Business processes underpin enterprise execution, coordination, and management. However, differing levels of familiarity with modeling languages among users can create an understanding gap, potentially disrupting the process flow. Business process documentation bridges this gap. Current methods, such as manual writing and rule-based generation, face inefficiency, errors, and limitations. We innovate by harnessing large language models for documentation generation. Our approach involves defining a Refined Process Structure Tree (RPST) meta-model and mapping rules, then constructing fine-grained RPSTs and crafting sentences using a hierarchical construction method. Finally, global optimization enhances the documentation. Tested on 100 diverse process models, our method outperforms benchmarks in robustness, and it achieves 6% and 1% higher semantic similarity scores by n-gram and semantics metrics.