Authors

Abstract

To address the issues of slow convergence in traditional methods, this study integrates adaptive gradient correction with a belief update mechanism based on the AdaBelief algorithm, and proposes a multimodal scenario control model for low-temperature grain storage. Experimental results indicate that the proposed model can maintain the grain storage temperature at approximately 12°C under various environmental conditions, and the average conductivity of wheat, rice, corn, and soybean is kept below 45 μs.cm-1. These findings demonstrate that the proposed model significantly improves the level of intelligent control in low-temperature grain storage systems and provides a novel approach for the precise regulation of grain storage environments.

Abstract in Chinese

低温粮食储存是一种环保技术，通过调节储存环境温度来减缓粮食代谢、抑制害虫繁殖和抑制霉菌生长。该方法正从单一目标温度控制向多目标智能调节演进，为全球粮食安全提供关键支持。为解决传统方法中收敛速度慢、参数调整滞后等问题，本研究基于AdaBelief算法提出了一种适用于低温粮食储存的多模态场景控制模型。该模型通过AdaBelief算法将自适应梯度校正与信念更新机制相结合，在粮食储存环境的多模态控制场景中展现出显著优势。实验结果表明，所提模型可在各种环境条件下将粮食储存温度维持在约12°C。不同谷物的导热系数均低于55，且该模型对害虫侵染也具有有效控制作用。该模型在温度控制方面实现了99.8%的高准确率，并在全年不同时段的温度控制性能上持续优于其他三种模型。这些研究结果表明，所提出的模型显著提升了低温粮食储存系统的智能控制水平，并为粮食储存环境的精准调控提供了新思路。