Junyu et al. (2026) Closed-loop autonomous scheduling of multi-level irrigation canal-gate systems using an LLM-agent framework: prompting strategies and model heterogeneity

Identification

• Journal: Computers and Electronics in Agriculture
• Year: 2026
• Date: 2026-09-22
• Authors: Chen Junyu, Wang Haiyu, Junzeng Xu, Zhang Binyang, Haoyang Zhang, Ye ZiJian, Zhang Zhonglili, Liu Xiaoyin, Rong Chen
• DOI: 10.1016/j.compag.2026.112444

Research Groups

• The State Key Laboratory of Water Disaster Prevention, Hohai University
• Jiangsu Province Engineering Research Center for Agricultural Soil-Water Efficient Utilization, Carbon Sequestration and Emission Reduction
• College of Agricultural Science and Engineering, Jiangning Campus, Hohai University
• Intelligent Equipment Research Center, Beijing Academy of Agriculture and Forestry Sciences

Short Summary

This study developed an LLM-based autonomous irrigation scheduling framework to address real-time control challenges in complex multi-level irrigation systems. The framework achieved reliable hydraulic regulation using a Constraint-Augmented (CA) prompting strategy.

Objective

• Investigate the application of Large Language Models (LLMs) in autonomous irrigation scheduling for multi-level canal-gate systems

Study Configuration

• Spatial Scale: A tree-topology irrigation network consisting of one main canal, three branch canals, and 60 irrigation fields.
• Temporal Scale: Four-hour continuous scheduling experiment

Methodology and Data

• Models used: GPT-5, DeepSeek-R1, Qwen3, Kimi-k2
• Data sources: Storm Water Management Model (SWMM) as a physics-driven hydraulic simulation environment for feedback generation, scheduling validation, and low-cost iterative debugging.

Main Results

• The Constraint-Augmented (CA) strategy effectively mitigated numerical insensitivity through embedded physical constraints.
• Qwen3-32B achieved a Stability Index (SI) of 99.7%, a Goal Achievement Rate (GAR) of 95%, zero constraint violations, and an irrigation uniformity coefficient of variation of 6%.

Contributions

• The study provides a pathway toward deployable cyber-physical irrigation scheduling systems by integrating LLM reasoning, hydraulic simulation, and MQTT-based communication.
• The results highlight the practical trade-off among regulation performance, inference efficiency, and deployment cost.

Funding

• This research was funded by the State Key Laboratory of Water Disaster Prevention (Grant No. SKLWD2023-01) and the Jiangsu Province Engineering Research Center for Agricultural Soil-Water Efficient Utilization, Carbon Sequestration and Emission Reduction (Grant No. JSERCA2024-02).

Citation

@article{Junyu2026Closedloop,
  author = {Junyu, Chen and Haiyu, Wang and Xu, Junzeng and Binyang, Zhang and Zhang, Haoyang and ZiJian, Ye and Zhonglili, Zhang and Xiaoyin, Liu and Chen, Rong},
  title = {Closed-loop autonomous scheduling of multi-level irrigation canal-gate systems using an LLM-agent framework: prompting strategies and model heterogeneity},
  journal = {Computers and Electronics in Agriculture},
  year = {2026},
  doi = {10.1016/j.compag.2026.112444},
  url = {https://doi.org/10.1016/j.compag.2026.112444}
}