AgriAgent: A Heterogeneous Multi-Agent Orchestration Framework for Autonomous, Knowledge-Grounded Precision Agriculture
提出多智能体框架提升精准农业自主性，数据详实，但属学术前沿，产业影响尚待验证。
面向精准农业受限于多模遥测难以实时自治的认知瓶颈,提出 AgriAgent 分层多智能体(MAS)框架,将大语言模型从被动分析工具重塑为目标导向的协调器,采用包含感知、认知、执行三层的三层编排(TLO)拓扑。构建基于 APSIM 生物物理模型并经 GAN 扩展的高保真 10,000 状态合成库,引入黑天鹅环境胁迫。基准测试显示推理成功率 98.2%,决策时延降低 97%,将传统人在回路模式转为智能体驱动,为生物-数字自治农场提供蓝图。
Open Access
| Issue | | EPJ Web Conf. Volume383, 2026 The 1 st International Conference on Artificial Intelligence and Smart Technologies for Green Energy (AIST4GE’26) |
| --- |
| |
| |
| Article Number | | 06004 |
| Number of page(s) | | 9 |
| Section | | Artificial Intelligence for Precision Agriculture and Hydraulic Systems |
| DOI | | https://doi.org/10.1051/epjconf/202638306004 |
| Published online | | 12 August 2026 |
EPJ Web of Conferences 383, 06004 (2026)
https://doi.org/10.1051/epjconf/202638306004
, Mohammed Berrahal*, Mimoune Yandouzi, Mounir Grari, Aissa Kerkour Elmiad and Mohammed Badaoui
IIA Research Lab, ESTO, Mohamed First University, Oujda, Morocco
- e-mail: [email protected]
** e-mail: [email protected]
Published online: 12 August 2026
Abstract
The global agricultural sector is currently navigating an unprecedented convergence of challenges, characterized by escalating caloric demands, systemic climate instability, and the urgent necessity for resource-neutral production. While the advent of Precision Agriculture (PA) introduced data-driven monitoring, the industry remains hindered by a "cognitive bottleneck" the inability to transform fragmented, multi-modal telemetry into real-time, autonomous interventions with-out intensive human oversight. This paper introduces AgriAgent, a pioneering, hierarchical Multi-Agent System (MAS) framework that redefines the role of Large Language Models (LLMs) from passive analytical tools to active, goal-oriented orchestrators. Unlike traditional monolithic architectures, AgriAgent em-ploys a Tri-Layer Orchestration (TLO) topology, comprising specialized agents for Perception, Cognition, and Actuation. To validate the framework's reasoning plasticity, we developed a high-fidelity synthetic library of N = 10,000 biological states, grounded in AP-SIM biophysical modeling and expanded via Generative Adversarial Networks (GANs) to include "Black Swan" environmental stressors. Our results demonstrate that by shifting from a reactive "human-in-the-loop" model to a proactive "agent-driven" paradigm, the system achieves a state of Ag-ronomic Equilibrium. Quantitative benchmarking shows a 98.2% reasoning suc-cess rate, a 97% reduction in decision latency, and a significant optimization of computational overhead. This research provides the definitive blueprint for the transition toward "Bio-Digital" autonomy, where the farm functions as a self-correcting, intelligent ecosystem capable of sustaining both global yield and eco-logical integrity in an increasingly volatile climate.
© The Authors, published by EDP Sciences, 2026
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.