Intelligent extraction of reservoir dispatching information integrating large language model and structured prompts

Area of Science:

• Hydrology and Water Resources Engineering
• Artificial Intelligence in Environmental Science
• Information Extraction and Natural Language Processing

Background:

• Reservoir dispatching regulations are vital for operational efficiency and decision-making.
• Current information extraction methods are time-consuming due to manual data labeling.
• Existing AI approaches for reservoir dispatch face challenges in cognitive load and output stability.

Purpose of the Study:

• To develop an efficient and accurate entity and relationship extraction method for reservoir dispatch using structured prompts.
• To create an AI agent that assists water resource professionals in acquiring structured data.
• To address the limitations of manual labeling and natural language prompt instability in AI-driven extraction.

Main Methods:

• Refining labels and organizing them using the Backus-Naur Form (BNF) to create a structured prompt format.
• Developing an AI agent guided by the structured prompt language for reservoir dispatch extraction.
• Conducting experimental verification to evaluate the AI agent's performance and efficiency.

Main Results:

• The AI agent effectively reduces cognitive burden and output instability for users.
• Achieved high extraction performance with F1 scores above 80% for both entities and relationships.
• Demonstrated the efficacy of structured prompt language in guiding large language models.

Conclusions:

• The proposed method offers a novel and effective solution for knowledge extraction in reservoir dispatch.
• This approach can be extended to other water resource management fields.
• The AI agent provides a practical tool for professionals, enhancing decision-making and intelligent recommendations.