Fu et al. (2026) Mitigating Peak Edge Effects in Multi-Zone Irrigation: A Safety-Constrained Reinforcement Learning Approach with Short-Term Evapotranspiration Forecasting

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2026
• Date: 2026-04-21
• Authors: Zhenyu Fu, Chunming Zhang, Xinwei Liu, Jihui Tian, Yu Song
• DOI: 10.3390/w18080988

Research Groups

Not specified

Short Summary

The study proposes a collaborative scheduling framework combining short-term evapotranspiration (ET) forecasting with safety-constrained reinforcement learning to optimize multi-zone campus irrigation. The approach successfully reduces water consumption and peak flow while eliminating constraint violations.

Objective

• To address peak edge operation and excessive valve switching in hydraulically coupled multi-zone irrigation systems while maintaining water efficiency and operational safety.

Study Configuration

• Spatial Scale: Multi-zone campus irrigation system.
• Temporal Scale: 7 consecutive days of field data (October 2025), utilizing 2-hour-ahead forecasting.

Methodology and Data

• Models used: Lightweight ET predictor, safety-constrained Reinforcement Learning (RL) incorporating a safety layer (Top-2 gating and total flow projection).
• Data sources: Field data including temperature, relative humidity, and light intensity (used to derive vapor pressure deficit and radiation proxy features).

Main Results

• Water Efficiency: Total water consumption was reduced to 131.04 m³, representing a decrease of 9.13% compared to fixed-schedule irrigation and 6.12% compared to hysteresis threshold rotational irrigation.
• Hydraulic Performance: Maximum total flow was reduced from 2.00 L/s to 1.60 L/s.
• Operational Stability: Valve switching cycles were lowered to 12, and border ratios at 0.90 and 0.95 were reduced to 0.
• Robustness: Ablation and noise/packet loss experiments confirmed that ET forecasting enables anticipatory scheduling and the safety projection layer ensures zero-violation operation.

Contributions

• Developed a practical and deployable scheduling framework that integrates predictive ET data with a safety-constrained RL policy to manage shared pump constraints in multi-zone irrigation.

Funding

Not specified

Citation

@article{Fu2026Mitigating,
  author = {Fu, Zhenyu and Zhang, Chunming and Liu, Xinwei and Tian, Jihui and Song, Yu},
  title = {Mitigating Peak Edge Effects in Multi-Zone Irrigation: A Safety-Constrained Reinforcement Learning Approach with Short-Term Evapotranspiration Forecasting},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18080988},
  url = {https://doi.org/10.3390/w18080988}
}