Author(s): Zhang Juan; Yang Moyuan; Shen Jianming; Zheng Fandong; Xu Zhilan; Li Xiaolin
Linked Author(s):
Keywords: Water resources scheduling; Water network knowledge graph; Intelligent decision-making for water networks; Multi-objective optimization; Large language models (LLMs)
Abstract: To address the trade-off between accuracy, efficiency, and interpretability in complex water network scheduling, this paper proposes a knowledge-driven multi-objective optimal scheduling model. The model integrates water network topology, operational rules, runoff mechanisms, and expert knowledge into a multi-modal knowledge base. Using graph-based reasoning and reinforcement learning, it enables autonomous path identification, scenario generation, and global optimization. A four-layer architecture is designed, coupled with large language models to support natural language-driven interactive decisions. Application results show: (1) overall benefits of the Digital Twin Yongding River project increased by >20%; (2) water source allocation rationality improved by 7%, and emergency response time reduced from 16 to 10 hours in the South-to-North Water Diversion project. The proposed model overcomes low efficiency and process fragmentation of traditional methods, offering reliable support for intelligent scheduling of complex water networks.
Year: 2026