AI-Assisted Pump Operation for Energy-Efficient Water Distribution Systems †

Abstract

Pumping water in water networks is generally the top energy demand for water systems. This study seeks to develop a large language model (LLM)-assisted framework for pump operation. Herein, ChatGPT was used to suggest pump control settings over 24 h that minimize energy use while maintaining pressure levels. In the proposed prompts, hourly information about the planned operation, i.e., pump control settings, minimum pressure levels, tank storage levels, and pump energy use, was provided. As the LLM suggests improved scenarios, EPANET results for these scenarios are fed back to it. This allows the LLM to learn and adjust future suggestions. The framework was validated on the example EPANET Net 3. Through iterative data exchange between the LLM and EPANET, the framework led to more energy-efficient pump scheduling. The LLM-assisted framework was compared with a genetic algorithm optimization. The results demonstrated that the proposed method outperformed the GA, achieving an energy reduction of 66.98%.