Existing retrieval-augmented generation (RAG) systems face challenges when applied to tasks such as elevator design computation workflow generation, which require strong domain expertise and high logical rigor. Specifically, guiding large language models through structured, step-by-step reasoning from complex and constantly updated design specifications remains a challenge. To address this issue, this paper proposes the ERPC-Prompt, a prompt optimization method based on explicit multi-step reasoning paths. The method aims to explicitly construct and optimize multi-step reasoning paths, while designing path-aware prompts to enable precise control over the reasoning process of large language models, thereby effectively improving the quality and reliability of generated design computation workflows. The specific research areas include: (1) how to explicitly construct multi-step reasoning paths tailored to the characteristics of elevator design computation workflows, and (2) how to use the constructed reasoning path information to optimize prompts that guide large language models in generating accurate, complete, and logically coherent step-by-step answers. Comparative experiments on the general-purpose logical reasoning dataset Multi-LogiEval and the custom-built elevator design computation workflow dataset validate the effectiveness of this method in enhancing the quality of design computation workflow generation by large language models.