AgriAgent: 面向农业环境控制的端到端大模型智能体系统架构

欢迎您访问《智慧农业（中英文）》官方网站！English

Toggle navigation智慧农业

• 网站首页
• 期刊介绍
• 编委会
• 投稿指南
• 期刊服务
• 下载中心
• 联系我们

Smart Agriculture ›› 2026, Vol. 8 ›› Issue (2): 220-236.doi: 10.12133/j.smartag.SA202507042

• 智能装备与系统 •上一篇下一篇

AgriAgent：面向农业环境控制的端到端大模型智能体系统架构

裘佳颖 1, 刘应昌 1, 高星杰 1, 黄远 2, 张红雨 1, 田芳 1,3, 李万理 1(), 冯在文 1,3,4()

1. 1. 华中农业大学信息学院，湖北 武汉 430070，中国

2. 华中农业大学园艺林学学院，湖北 武汉 430070，中国

3. 云南现代农业产业研究院有限公司，云南 昆明 650032，中国

4. 农业智能技术教育部工程研究中心，湖北 武汉 430070，中国

• 收稿日期:2025-07-30 出版日期:2026-03-30
• 基金项目:湖北省技术创新计划项目(2024BBB055); 湖北省科技厅援藏项目(2024EIA009); 云南省重大科技专项计划(202502AE090003); 中央高校基本科研业务费专项资金(2662025XXPY005); 中央高校基本科研业务费专项资金(2662023XXQD002); 农业融合出版知识挖掘与知识服务重点实验室开放课题(2025KMKS01)
• 作者简介:裘佳颖，本科，研究方向为自然语言处理。E-mail：[email protected]； 刘应昌，硕士研究生，研究方向为自然语言处理。E-mail：[email protected] 裘佳颖，本科，研究方向为自然语言处理。E-mail：[email protected]； 裘佳颖和刘应昌并列第一作者
• 通信作者:李万理，博士，讲师，研究方向为自然语言处理、知识图谱、元学习。E-mail：[email protected]； 冯在文，博士，副教授，研究方向为农业AI大模型。E-mail：[email protected]

AgriAgent: End-to-End Large Model Agent System Architecture for Agricultural Environment Control

QIU Jiaying 1, LIU Yingchang 1, GAO Xingjie 1, HUANG Yuan 2, ZHANG Hongyu 1, TIAN Fang 1,3, LI Wanli 1(), FENG Zaiwen 1,3,4()

1. 1. College of Informatics, Huazhong Agricultural University, Wuhan 430070, China

2. College of Horticulture and Forestry, Huazhong Agricultural University, Wuhan 430070, China

3. Yunnan Modern Agricultural Industry Research Institute, Kunming 650032, China

4. Engineering Research Center of Agricultural Intelligent Technology, Ministry of Education, Wuhan 430070, China

• Received:2025-07-30 Online:2026-03-30
• Foundation items:Hubei Provincial Technology Innovation Plan Project(2024BBB055); Tibet Assistance Project of Department of Science and Technology of Hubei Province(2024EIA009); Yunnan Provincial Major Science and Technology Special Project Plan(202502AE090003); Fundamental Research Funds for the Chinese Central Universities(2662025XXPY005); Open Project of Key Laboratory of Knowledge Mining and Knowledge Service for Agricultural Integrated Publishing(2025KMKS01)
• About author:QIU Jiaying, E-mail: [email protected]; LIU Yingchang, E-mail: [email protected]
• Corresponding author:LI Wanli, E-mail: [email protected]; FENG Zaiwen, E-mail: [email protected]

在线阅读 99

知网下载

本地下载 86

## 可视化 0

摘要/Abstract

摘要：

【目的/意义】 为突破大语言模型在农业生产决策领域应用中存在的物理世界交互能力不足、专业工具调用受限，以及“幻觉”生成等问题，构建了一个工具增强型环控智能体系统。 【方法】 首先，基于开放气象数据与作物生长模型构建农业数字孪生评测平台，提供标准化闭环控制验证环境。其次，提出了AgriAgent架构，通过“传感器-记忆库-检索器-大语言模型-工具执行器”闭环流程，实现了田间环境的动态感知、知识检索、推理决策与精准执行的端到端闭环控制。该端到端架构的优势在于实现了“环境感知-动态决策-执行反馈”的全流程无缝衔接，极大地提升了系统在复杂、动态农业环境下的实时响应能力和决策精度，解决了传统应用缺乏“感知-决策-执行”闭环能力的问题。为验证系统有效性，在数字孪生环境中，对玉米、小米、甜菜、番茄、卷心菜5种作物在常规、扰动、极端3种情景下开展实验。 【结果和讨论】 以Qwen2.5_7B为核心的AgriAgent-7B模型表现最优，常规情景、扰动情景和极端情景下玉米产量较基线模型分别提升562.20%、459.70%、463.60%；小米产量分别提升646.80%、558.70%、351.20%；甜菜产量分别提升228.20%、118.20%、125.40%；番茄产量分别提升926.20%、979.55%、1 537.46%；卷心菜产量分别提升501.67%、504.65%、1 185.14%。此外，智能体参数量显著影响决策的质量，当参数量达到7 B时，智能体展现了卓越的灵活性与适应性，能够解决大部分复杂的农业生产决策问题。 【结论】 本研究验证了工具增强型智能体在农业动态决策中的有效性，提出的AgriAgent端到端架构，通过知识检索与工具调用闭环，在农业场景下实现了“环境感知-动态决策-执行反馈”的全周期闭环控制，为农业智能体研究提供了可靠技术路径和创新范式。

关键词:工具增强型大语言模型,农业智能体,数字孪生,农业评测平台,DSSAT/作物生长模型

Abstract:

[Objective] Large language models (LLMs) have demonstrated strong capabilities in natural language understanding, knowledge integration, and complex reasoning, offering new opportunities for intelligent decision-making in agriculture. However, their direct application in agricultural production and facility environment control remains challenging due to strong physical constraints and high operational risks. The lack of real-world interaction and executable decision grounding limits the practical effectiveness of conventional LLMs in such scenarios. To address these challenges, a tool-augmented LLM-based agricultural intelligent agent system, termed AgriAgent, was proposed, and a digital-twin-based evaluation platform for agricultural decision-making was developed. By integrating a high-fidelity digital twin environment with an end-to-end agent architecture, the decision-making performance of agricultural intelligent agents with different parameter scales was systematically evaluated across multiple crops and climate scenarios. [Methods] A high-fidelity agricultural digital twin evaluation platform was constructed using the decision support system for agrotechnology transfer (DSSAT) v4.8 crop growth model as the core simulation engine to model crop growth under diverse environmental conditions and management strategies. Meteorological driving data were obtained from the Seoul Historical Weather Data dataset. Through data cleaning, missing-value imputation, unit normalization, and time-series reconstruction, the raw meteorological data were transformed into standardized inputs compatible with DSSAT. Three climate scenarios representing different environmental complexities were designed, including a regular scenario, a perturbed scenario, and an extreme scenario. The regular scenario employed historical observations, the perturbed scenario introduced stochastic disturbances to simulate short-term climate variability, and the extreme scenario incorporated multi-factor coupled stresses such as high temperatures and excessive precipitation during sensitive growth stages. In total, 90 annual climate driving sequences were generated. Fixed soil profile parameters calibrated by domain experts were applied across all simulations to minimize confounding effects. Within this digital twin environment, a tool-augmented agricultural intelligent agent, AgriAgent, was implemented using a modular architecture consisting of a sensor module, memory module, retriever, large language model, and tool executor, forming a closed-loop decision-making framework. In each decision cycle, the agent perceived environmental and crop state information, including soil moisture and nutrient status, meteorological conditions, crop growth stages, and stress indicators. State summaries and historical decisions were stored in memory, while agronomic knowledge was retrieved through a retrieval-augmented generation mechanism. Based on integrated information, the LLM generated structured environmental control commands in JSON format, which were validated and constrained by the tool executor before updating the DSSAT environment. The system supported irrigation, supplementary lighting, ventilation, heating, fertilization, and CO 2 enrichment. Five representative crops: maize, millet, sugar beet, tomato, and cabbage were simulated under the three climate scenarios over complete growing seasons, resulting in 450 crop-scenario combinations. An unmanaged DSSAT simulation served as the baseline. AgriAgent models with three parameter scales (1.5B, 3B, and 7B), built on the Qwen2.5 series, were evaluated. Crop economic yield expressed as dry matter at physiological maturity was adopted as the evaluation metric. [Results and Discussions] The results showed that AgriAgent consistently outperformed the baseline across all crops and climate scenarios, with model scale exerting a significant influence on decision-making performance. AgriAgent-7B achieved the best overall performance under regular, perturbed, and extreme scenarios, demonstrating strong generalization ability and environmental adaptability. By dynamically adjusting water, nutrient, light, and thermal management strategies, the agent effectively mitigated environmental stresses even under multi-factor coupled extreme climate conditions. Under extreme scenarios, AgriAgent-7B increased yields by 463.60% for maize, 351.20% for millet, 125.40% for sugar beet, 1 537.46% for tomato, and 1 185.14% for cabbage compared with the baseline. Particularly large gains were observed for high-value crops such as tomato and cabbage, highlighting the advantages of the proposed framework for precision-controlled facility agriculture. In contrast, AgriAgent-1.5B exhibited performance comparable to the baseline, while AgriAgent-3B achieved moderate improvements but remained inferior to the 7B model. These findings indicate a clear scaling effect, suggesting that larger models possess stronger capabilities in multi-source information integration, long-term temporal reasoning, and adaptation to complex environments. [Conclusions] This study developed a digital-twin-based agricultural decision evaluation platform and proposed a tool-augmented, end-to-end agricultural intelligent agent named AgriAgent. Experiments across multiple crops and climate scenarios verified the effectiveness and robustness of the proposed framework for dynamic agricultural decision-making. The results demonstrate that integrating knowledge retrieval, reasoning, and tool execution within a closed-loop LLM-based agent enables stable, reliable, and adaptive environmental control, providing a feasible technical pathway and standardized evaluation paradigm for intelligent agriculture.

Key words:tool-augmented large language model,agricultural agent,digital twin,agricultural evaluation platform,DSSAT/crop growth model

中图分类号:

• S126

• TP18

引用本文

裘佳颖, 刘应昌, 高星杰, 黄远, 张红雨, 田芳, 李万理, 冯在文. AgriAgent：面向农业环境控制的端到端大模型智能体系统架构[J]. 智慧农业(中英文), 2026, 8(2): 220-236.

QIU Jiaying, LIU Yingchang, GAO Xingjie, HUANG Yuan, ZHANG Hongyu, TIAN Fang, LI Wanli, FENG Zaiwen. AgriAgent: End-to-End Large Model Agent System Architecture for Agricultural Environment Control[J]. Smart Agriculture, 2026, 8(2): 220-236.

使用本文

/[推荐](javascript:void(null))
导出引用管理器EndNote|Ris|BibTeX

链接本文:https://www.smartag.net.cn/CN/10.12133/j.smartag.SA202507042

https://www.smartag.net.cn/CN/Y2026/V8/I2/220

[图/表 14](javascript:;)

图1

工具增强型环控智能体系统架构示意图

图1

表1

AgriAgent控制系统建模

设备	控制动作符号	范围/分辨率	主要影响
灌溉系统	I	0~30 mm/h	提升 W s o i l$W_{s o i l}$
补光系统	L	0~200%	提升光合有效辐射
通风/加温	V，H	0/1（开关）	调节 T a v g,W S$T_{a v g} , W S$
施肥系统	F	0~50 kg/hm 2	增加 N t o t a l$N_{t o t a l}$​
CO 2 施放	C	0~800 μmol/mol/h	提升室内CO₂

表1

图2

数字孪生农场功能示意图

图2

图3

AgriAgent架构示意图

图3

表2

AgriAgent系统组件核心职能与关键实现要点

模块	功能摘要	实现要点
传感器	从仿真环境读取状态向量 s t$\mathbf{\mathit{s}}_{t}$	以API形式模拟真实传感器
记忆库	持久化存储全局状态摘要	以文本形式保存在System prompt中
格式转换器	将结构化状态拼接为自然语言片段	以固定规则实现
检索器（Retriever）	面向知识库检索作物生理模型、控制经验	以Faiss检索实现
LLM	解析环境信息并生成JSON格式工具调用	经过预训练和后训练的大语言模型
工具执行器	校验、裁剪并调用修改DSSAT 的接口	Python实现

表2

表3

数字孪生农场土壤的理化性质与特征参数

土层深度/cm	［0，20）	［20，40）	［40，60）	［60，80）	［80，100）
颗粒组成/%	黏粒	21.6	21.4	20.8	11.2
粉粒	78.4	78.6	78	78.1	69.3
砂粒	0.0	0.0	1.2	10.7	20.3
土壤容重/（g/cm 3）	1.47	1.44	1.52	1.54	1.46
田间持水量/（cm 3/cm 3）	0.306 0	0.274 9	0.319 1	0.375 6	0.328 8
永久凋萎点/（cm 3/cm 3）	0.080 6	0.095 2	0.093 1	0.072 3	0.093 1
土壤有机碳/%	1.0	0.5	0.5	0.4	0.3
pH	8.47	8.80	8.82	8.84	8.85
饱和含水量/（cm 3/cm 3）	0.410 5	0.420 9	0.257 8	0.426 0	0.429 2
饱和导水率/（cm/d）	45.60	25.13	55.56	7.49	130.30

表3

图4

AgriAgent模型（1.5B， 3B， 7B）与基线模型在5种作物和3种情景下的性能对比（以最终作物干物质产量（kg/hm2）衡量，数据为30个独立气候轨迹的平均值）

图4

表4

不同参数量AgriAgent模型在5种作物与3种情景下的平均最终产量干重 ( kg/hm2)

作物	模型	常规	扰动	极端
玉米	Baseline	990.10±144.55	997.60±142.00	600.67±524.62
AgriAgent-1.5B	990.10±144.55	997.60±142.00	600.67±524.62
AgriAgent-3B	5 440.80±1 520.62	5 287.33±1 471.01	3 304.53±2 950.12
AgriAgent-7B	6 557.83±950.12	5 584.17±1 200.45	3 385.20±2 900.34
小米	Baseline	867.20±268.49	894.63±268.79	721.97±426.74
AgriAgent-1.5B	867.20±268.49	894.63±268.79	721.97±426.74
AgriAgent-3B	5 281.03±1 174.62	5 291.87±1 428.79	3 431.40±1 948.76
AgriAgent-7B	6 474.03±1 400.56	5 892.70±1 200.78	3 257.83±1 800.45
甜菜	Baseline	8 587.90±6 250.11	8 853.70±6 459.85	8 845.07±7 099.96
AgriAgent-1.5B	8 592.87±6 265.19	7 835.5±6 465.73	9 081.30±7 113.27
AgriAgent-3B	16 755.30±5 911.93	13 818.70±6 247.26	15 365.40±7 188.85
AgriAgent-7B	28 186.00±2 745.67	19 315.33±2 400.50	19 938.23±2 800.25
番茄	Baseline	39.43±57.93	35.20±135.13	22.13±91.36
AgriAgent-1.5B	39.43±57.93	35.20±135.13	22.13±91.36
AgriAgent-3B	367.97±178.01	352.67±142.79	300.03±194.80
AgriAgent-7B	404.63±120.34	380.00±150.67	362.17±200.12
卷心菜	Baseline	88.87±6.52	84.97±7.32	35.20±25.99
AgriAgent-1.5B	88.87±6.52	84.97±7.32	35.20±25.99
AgriAgent-3B	398.53±127.86	336.60±158.31	311.57±142.41
AgriAgent-7B	534.70±80.12	513.77±70.45	452.27±90.23

表4

图5

AgriAgent-7B模型调控下玉米、小米和甜菜收获物干重随时间的变化曲线（常规情景）

图5

图6

AgriAgent 在设施环境控制中的“感知-决策-执行”闭环流程样例：基于实时环境数据的工具调用决策

图6

表5

高影响幻觉（权重1.0）类型及核心特征：决策错误、天气条件与产量影响关联

决策错误	天气条件	对产量的影响
极端高温下灌溉不足	温度>作物临界高温	作物热胁迫，光合作用受阻，直接减产
极端高温下大量施肥	温度>作物临界高温+5 ℃	烧苗，根系损伤，不可逆伤害
干旱条件灌溉不足	降水少+温度高	水分胁迫，生长停滞，严重减产
低湿度高温灌溉不足	湿度低+温度高	过度蒸