基于物联网与大语言模型的农作物生长在线诊断与决策系统

Online Diagnosis and Decision-making System for Crop Growth Based on Internet of Things and Large Language Models

摘要:

随着市场对农产品精细化与多样化需求的持续增长,现代农业的发展重心已从单一追求产量转向了品质提升,对智能化、标准化种植技术的需求日益迫切,然而,现有农作物生长诊断系统仍受限于作物生长特征综合表征能力不足与先进种植技术向农户推广困难等关键技术瓶颈。针对这一关键问题,构建了一套农作物生长智能诊断与决策系统。通过精准记录农作物的性状特征及传感器实时监测的环境参数,建立了理想生长模板与实际作物的生长模型。通过对比理想生长模板与实际作物的差异特征,捕捉生长模式的关键区别,同时引入大模型GLM-4-Plus,综合其他种植情况及现存问题构建大模型提示词,为农业种植提供针对性的反馈与优化建议并通过现场控制层执行调控。为验证该系统的实际应用效果,在山东省潍坊市的大棚农作物种植环境中进行系统部署。应用效果显示,该系统能够有效分析当前农作物的生长状况,提供科学的种植建议。

Abstract:

The integration of Internet of Things (IoT) technology with large language models (LLMs) offers transformative potential for precision agriculture. It introduced an intelligent crop growth diagnosis and decision-making system that combined IoT sensors with the GLM-4 large model to enhance agricultural efficiency and management. The system analyzed crop conditions to provide real-time agronomic recommendations, bridging the gap between advanced research and smallholder farmers. The system architecture comprised four layers: field control, system modeling, data analysis, and decision support. The field control layer used multi-source sensors to collect soil and air parameters, and crop phenotypes, uploading data via an ARM processor and 4G module. The system modeling layer integrated literature, expert knowledge, and real-time data to create crop growth and environmental response models. The data analysis layer employed temporal alignment algorithms to compare ideal templates with actual data, quantifying deviations and generating prompts for the large model. Leveraging the GLM-4-Plus model, it provided solutions for environmental regulation and pest/disease management, implemented through actuators for closed-loop control. The system's core innovation was its collaborative analysis framework combining “sensor-based quantitative monitoring + farmer natural language description + large model intelligent diagnosis”. This approach integrated quantitative sensor data with farmers' qualitative observations of crop abnormalities, overcoming traditional IoT limitations in detecting subtle growth anomalies. The domestically developed GLM-4-Plus was applied to an agricultural decision support system, using structured prompt processing for precise diagnosis of issues like nutrient deficiencies and environmental stresses, and targeted interventions such as irrigation adjustments and fertilizer optimization. A two-month field trial in Shouguang City, Shandong Province, validated the system's effectiveness across tomatoes, sweet peppers and strawberries. A case study demonstrated the system's ability to diagnose yellowing tomato leaves, attributing it to suboptimal temperature, low humidity, and insufficient light. Implementing GLM-4-recommended measures significantly improved plant health. The system also addressed uneven fruit development and excessive stem elongation, demonstrating adaptability across growth stages. Future research aimed to expand crop coverage, enhance environmental robustness, and explore edge computing for real-time control. By synergizing IoT-based sensing with LLM-driven decision-making, the system democratized advanced agronomic knowledge, offering a practical pathway for data-driven sustainable agriculture. Empirical results confirmed its potential as a key technology for global agricultural transformation.