通讯作者: 朱毅华, zhyh@njau.edu.cn; 汤亮, tangl@njau.edu.cn

DOI：10.12343/Agrixiv.202610.00002  CSTR：31252.36.agrixiv.202610.00002

SmartFarming: a GIS-based agent framework for farm management

• 摘要：[目的] 农场管理需要持续决策，并受田块条件、作物、气象、土壤、农业投入与经营目标的约束；然而基于预定义软件模块的工作流仍难以在平台与模型之间传递田块边界、变量语义、参数单位与中间结果。本研究构建了 SmartFarming——一个基于地理信息系统（GIS）的智能体框架，可将自然语言的农场管理需求转化为受约束、可执行、可追溯的多步工作流。[方法] 该框架由角色（Persona）、智能体运行时（Agent Harness）、能力（Capabilities）与世界接口（World Interface）四个核心部分组成。智能体运行时负责协调任务编排、能力路由、上下文管理、分层记忆、执行监控与治理；技能契约约束能力组合；产物及其清单记录空间语义、版本与数据谱系；GIS 模块在空间反馈闭环中连接田块输入、计算、结果可视化与用户复核。[结果] 以 9 个案例实例化该框架，涵盖卫星遥感监测、无人机遥感水稻生长监测、地面 RGB 图像的小麦穗与水稻穗计数及病害识别、作物生长预测，以及模型-遥感耦合产量预测。案例表明，SmartFarming 能够在端到端工作流中组织、组合并校验已有的农学模型、数据服务与空间交互。[结论] 具有显式空间、时间与风险约束的基于 GIS 的智能体框架，为农场管理中异构农学模型与数据服务的复用提供了可行基础，并为构建农业领域智能体系统提供了可迁移的结构。

  关键词： 农场管理; 智能体运行时; 技能契约; 地理信息系统; 遥感

   

  Abstract: Objective: Farm management requires continuous decisions constrained by field conditions, crops, weather, soil, agricultural inputs and business objectives, yet workflows built on predefined software modules still fail to transfer field boundaries, variable semantics, parameter units and intermediate outputs across platforms and models. This study develops SmartFarming, a GIS-based agent framework that translates natural-language farm management requests into constrained, executable and traceable multistep workflows.Methods: The framework is organized around four components - Persona, Agent Harness, Capabilities and World Interface. The Agent Harness coordinates task orchestration, capability routing, context management, hierarchical memory, execution monitoring and governance. Skill contracts constrain capability composition, while artifacts and their manifests record spatial semantics, versions and data lineage. A GIS module connects field inputs, computation, result visualization and user review within a spatial feedback loop.Results: Nine case studies instantiate the framework: satellite remote sensing monitoring, UAV remote sensing for rice growth, ground RGB image analysis for wheat-head and rice-panicle counting and disease identification, crop growth prediction, and model-remote-sensing coupling for yield prediction. The cases show that SmartFarming can organize, compose and check existing agronomic models, data services and spatial interactions in end-to-end workflows.Conclusions: A GIS-based agent framework with explicit spatial, temporal and risk constraints provides a practical basis for reusing heterogeneous agronomic models and data services in farm management, and offers a transferable structure for building domain agent systems in agriculture.

  Key words: farm management; agent harness; skill contract; geographic information system; remote sensing

  提交时间：2026-10-09

  版权声明：作者本人独立拥有该论文的版权，预印本系统仅拥有论文的永久保存权利。任何人未经允许不得重复使用。
• html

• 图表

• 朱俊卓, 陈铭, 张若雨菲, 穆再排尔・吐尔孙, 玉米提江・阿布杜热依木. 干旱区智慧农业智能水肥一体化系统设计与应用研究. 2026. doi: 10.12343/Agrixiv.202607.00002

  刘思嘉, 陈铭, 张若雨菲, 朱俊卓, 穆再排尔·吐尔孙. 基于土壤墒情预测的干旱区棉花智慧灌溉决策系统研究. 2026. doi: 10.12343/Agrixiv.202607.00006

  George Worrall, Jasmeet Judge, Kenneth J. Boote, Anand Rangaraja. In-season crop progress estimation using remote sensing and model-guided machine learning.. 2022. doi: 10.31220/agriRxiv.2022.00131

  郑建华, 赵瑞雪, 赵华, 李思经, 王国刚. 智慧农业信息服务发展的国际经验与启示. 2026. doi: 10.12343/Agrixiv.202604.00004

  Dyan Meuwissen, Kim Brosens, Ben Aernouts, Ines Adriaen. Sample frequency and reactivity of farmers influences correct insemination window on farms using on-farm milk progesterone for fertility management.. 2022. doi: 10.31220/agriRxiv.2022.00133

  张洁, 李甜, 赵华, 赵瑞雪, 寇远涛, 朱亮. 元宇宙赋能行业联盟资源共建共享：服务场景与发展路径 ——以国家农业大数据与信息服务联盟为例. 2026. doi: 10.12343/AgriXiv.202603.00002

  Anna H. Stygar, Lilli Frondelius, Greta Veronica Berteselli, Yamenah Gómez, Elisabetta Canali, Jarkko K. Niemi, Pol Llonch, Matti Pastel. Measuring dairy cow welfare with real-time sensor-based data and farm records: a concept study.. 2023. doi: 10.31220/agriRxiv.2023.00185

• 序号	提交日期	编号	操作
  1	2026-10-05	10.12343/Agrixiv.202610.00002V1	下载

• 公开评论  匿名评论  仅发给作者 提交评论