Key points are not available for this paper at this time.
Standards and specifications are vital to the high-quality construction and long-term safety of hydropower engineering. However, the traditional manual consultation pattern is inefficient and error-prone, potentially posing risks to engineering quality and safety. Therefore, promoting standard digitalisation is vital to supporting knowledge-informed decision-making and enhancing safety management. To achieve this goal, this paper constructs an intelligent Q&A system for hydropower engineering standards and specifications based on large language models (LLMs). As a core component of the system, a Chinese Specification Text Splitter (CSTS) is developed, which parses PDF using the MinerU tool and implements a tree-based splitting algorithm that respects the hierarchical structure of standard documents. The domain knowledge base constructed upon CSTS achieves unified representation of multimodal elements (texts, formulas, tables and images) while maintaining semantic integrity. Additionally, a benchmark comprising 1168 professional questions is established to enable systematic evaluation. Experiments demonstrate that the system, powered by CSTS, significantly improves response quality, with comprehensive scores increasing by 6% and 7% compared to the traditional Recursive Character Text Splitter (CRTS) and Markdown Header Text Splitter (MHTS), respectively. The knowledge Q&A system provides efficient and precise standards knowledge query services, which are important for advancing the standard digitalisation in hydropower engineering industry.
Liu et al. (Tue,) studied this question.