Scholarlink

<p>Efficient irrigation management is critical for sustainable agriculture, particularly in regions facing water scarcity and climate variability. Traditional irrigation practices lack real-time adaptability, often resulting in excessive water consumption and inconsistent soil moisture regulation. This research proposes an intelligent irrigation scheduling framework that integrates multi-source environmental data with advanced deep learning and reinforcement learning techniques. The system incorporates soil moisture sensor readings, weather parameters, and remote-sensing vegetation indices, which undergo normalization, feature engineering, and autoencoder-based representation learning. A hybrid Transformer–GRU architecture captures long-range temporal dependencies and short-term sequential patterns, while static soil and crop attributes are integrated through a feedforward network. Reinforcement learning enables adaptive irrigation control by selecting optimal water application strategies through a sustainability-aware reward function. The proposed system achieved high accuracy, with an MAE of 0.85, improved water usage efficiency by 35%, and increased crop yield to 6,200&#xa0;kg/hectare while reducing drought risk by 60%. Enhanced NDVI and stable soil moisture levels further validated model performance. Future work will focus on large-scale field deployment, edge-computing integration for real-time automation, and extension to multi-crop and climate-resilient agricultural systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing Irrigation Scheduling with a Hybrid Transformer-GRU Model and Reinforcement Learning in Smart Agriculture

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Shriya Sahu,
  • Prerna Verma

摘要

Efficient irrigation management is critical for sustainable agriculture, particularly in regions facing water scarcity and climate variability. Traditional irrigation practices lack real-time adaptability, often resulting in excessive water consumption and inconsistent soil moisture regulation. This research proposes an intelligent irrigation scheduling framework that integrates multi-source environmental data with advanced deep learning and reinforcement learning techniques. The system incorporates soil moisture sensor readings, weather parameters, and remote-sensing vegetation indices, which undergo normalization, feature engineering, and autoencoder-based representation learning. A hybrid Transformer–GRU architecture captures long-range temporal dependencies and short-term sequential patterns, while static soil and crop attributes are integrated through a feedforward network. Reinforcement learning enables adaptive irrigation control by selecting optimal water application strategies through a sustainability-aware reward function. The proposed system achieved high accuracy, with an MAE of 0.85, improved water usage efficiency by 35%, and increased crop yield to 6,200 kg/hectare while reducing drought risk by 60%. Enhanced NDVI and stable soil moisture levels further validated model performance. Future work will focus on large-scale field deployment, edge-computing integration for real-time automation, and extension to multi-crop and climate-resilient agricultural systems.