Automated SQL Query Generation for an Intelligent Sewer Management System Using Large Language Models

The aging U.S. wastewater infrastructure poses significant environmental and public health risks. While municipalities collect data for asset management, handling large datasets can overwhelm decision-makers. Natural language processing (NLP) simplifies data access by enabling users to retrieve information through voice or text commands, reducing reliance on IT support and improving management efficiency. This paper investigates automated SQL query generation from natural language, a critical step in developing an intelligent sewer management system. Large Language Models (LLMs) are employed to achieve it. The proposed method adopts several techniques to improve the performance of SQL query generation, including data augmentation, prompt engineering and fine-tuning. Experimental results are presented to evaluate the proposed method with different settings and compare it with a baseline method.