基于1D-CNN的稻谷石墨烯远红外干燥模型及含水率在线预测  

Graphene-based far-infrared drying model of rice grains and online moisture content prediction using 1D-CNN

作　　者：王逸凡 井世亮 夏宇 Nuhu Jibril 赵海瑞 陈坤杰[1] WANG Yifan;JING Shiliang;XIA Yu;Nuhu Jibril;ZHAO Hairui;CHEN Kunjie(College of Engineering,Nanjing Agricultural University,Nanjing 210031,China;Jiangsu Agricultural Machinery Testing Station,Nanjing 210017,China)

机构地区：[1]南京农业大学工学院,江苏南京210031 [2]江苏省农业机械鉴定站,江苏南京210017

出　　处：《南京农业大学学报》2025年第2期488-497,共10页Journal of Nanjing Agricultural University

基　　金：江苏省科技计划专项资金(重点研发计划现代农业)项目(BE2021305)。

摘　　要：[目的]为实现远红外稻谷干燥过程中水分比的精准预测,提出了一种基于1D-CNN(one-dimensional convolutional neural network)的稻谷干燥水分比预测模型,实现对干燥过程中稻谷含水率在线预测。[方法]将稻谷初始含水率调到统一标准后,在自制的石墨烯远红外干燥试验台进行不同温度的干燥试验,每隔2 min采集1组包括干燥温湿度等8个工艺参数数据,经标准化处理后构成数据集。然后以8个工艺参数为输入,水分比为输出,构建1D-CNN干燥模型,通过训练确定模型参数,最后对模型进行验证并与6种经典薄层干燥模型及4种典型机器学习干燥模型进行比较。[结果]试验结果表明,所提出的1D-CNN干燥模型能够很好描述干燥过程中水分比的变化情况,决定系数R^(2)、均方根误差(root mean square error, RMSE)和平均绝对误差(mean absolute error, MAE)分别达到0.993 1、0.018 9、0.012 1;含水率预测的MAE和平均相对误差(mean relative error, MRE)分别为0.143 2%和0.007 8%,明显优于其他对比的干燥模型。[结论]所提出的1D-CNN干燥模型能够准确预测稻谷干燥过程中含水率变化,完全满足含水率在线检测需求。[Objectives]To achieve precise prediction of the moisture ratio during the far-infrared drying process of rice grains,a rice drying moisture ratio prediction model based on one-dimensional convolutional neural network(1D-CNN)had been proposed,enabling online prediction of the moisture content of rice grains during drying.[Methods]After standardizing the initial moisture content of rice grains,drying experiments at various temperatures were conducted on a homemade graphene far-infrared drying experimental platform.A set of eight process parameter data,including drying temperature and humidity,was collected every 2 minutes and subjected to normalization treatment to form the dataset.The 1D-CNN drying model was constructed with the eight process parameters as inputs and the moisture ratio as output.The model parameters were determined through training,and the model was validated and compared with six classic thin-layer drying models and four typical machine learning drying models.[Results]Experimental results demonstrated that the proposed 1D-CNN drying model was capable of accurately depicting the variation of moisture ratio during the drying process.The determination coefficient(R^(2)),root mean square error(RMSE),and mean absolute error(MAE)were achieved at 0.9931,0.0189,and 0.0121,respectively.The MAE and mean relative error(MRE)for moisture content prediction were 0.1432%and 0.0078%,respectively,significantly outperforming the other compared drying models.[Conclusions]The proposed 1D-CNN drying model could accurately predict the changes in moisture content during the rice drying process,fully meeting the requirements for online moisture content detection.

关 键 词：稻谷 1D-CNN模型 石墨烯远红外干燥 含水率在线预测 

分 类 号：S375[农业科学—农产品加工] S226.6[农业科学—农艺学]