一种仓储粮堆温湿度场空间高分辨率重构模型  

A Spatial High-Resolution Reconstruction Model for Warehouse Grain Piles Temperature and Humidity Fields

作　　者：曹云皓 段珊珊[1] 杨卫东[2] 白春启[3] 赵超 万安平 Cao Yunhao;Duan Shanshan;Yang Weidong;Bai Chunqi;Zhao Chao;Wan Anping(College of Information Science and Engineering,Henan University of Technology,Zhengzhou 450001;School of Artificial Intelligence and Big Data,Henan University of Technology,Zhengzhou 450001;School of Food and Strategic Reserves,Henan University of Technology,Zhengzhou 450001;Grain Intelligence Technology Co.,Ltd.,Guangzhou 510665)

机构地区：[1]河南工业大学信息科学与工程学院,郑州450001 [2]河南工业大学人工智能与大数据学院,郑州450001 [3]河南工业大学粮食和物资储备学院,郑州450001 [4]广州谷物智能科技有限公司,广州510665

出　　处：《中国粮油学报》2024年第12期17-25,共9页Journal of the Chinese Cereals and Oils Association

基　　金：河南省重大公益专项项目(201300210100)

摘　　要：仓储粮堆内的温湿度数据是粮食安全储存、保质减损的核心指标。目前仓储粮堆内布设的温湿度测量点均为稀疏离散点,其他未布设传感器的区域的温度和湿度无法直接获取,生成温湿度场的方法主要基于插值法等数学模型,误差较大。本研究通过在实验仓的仓储粮堆中布设温湿度传感器网格以采集数据,并使用基于高分辨率重构的深度学习方法,提出了一种仓储粮堆的温湿度场空间重构模型及算法。实验结果表明,相较于传统的线性插值法,本研究的温湿度场重构模型效果更好,粮堆温度重构值与测量值的平均绝对误差(MAE)小于0.30℃,均方根误差(RMSE)小于0.35℃;粮堆湿度重构值与测量值的平均绝对误差(MAE)小于0.51%,均方根误差(RMSE)小于1.14%。Storage grain temperature and humidity data are crucial indicators for the safe storage and quality preservation of grains.Currently,the temperature and humidity measurement points deployed in storage grain piles are sparse and discrete.The temperature and humidity of areas without sensors cannot be directly obtained.Methods for generating temperature and humidity fields are mainly based on mathematical models such as interpolation,leading to significant errors.In this study,this issue was addressed by deploying a grid of temperature and humidity sensors in a test grain storage facility,collecting data,and proposing a spatial reconstruction model and algorithm for the temperature and humidity field in storage grain piles based on high-resolution reconstruction using deep learning.Experimental results indicated that,compared to traditional linear interpolation methods,the proposed temperature and humidity field reconstruction model in this study performed better.The average absolute error(MAE)between the predicted and measured values of grain pile temperature was less than 0.30℃,and the root mean square error(RMSE)was less than 0.35℃.For grain pile humidity,the MAE was less than 0.51%,and the RMSE was less than 1.14%.

关 键 词：深度学习 高分辨率重构 仓储粮堆 温湿度场空间重构 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]