摘要
为了实现玉米含水率的快速无损检测,该文利用精密阻抗分析仪和自制介电参数测量传感器通过激励电压在1 kHz^5.462 MHz频率范围内测量了热风干燥过程中不同含水率与不同温度下玉米籽粒的介电常数ε′和介电损耗ε″。通过对双介电参数频谱的分析,对含水率回归模型建模频段进行了初步选择,以1.072~5.462 MHz之间15个测量频点的双介电参数和温度值T共计31维变量作为支持向量回归机(support vector regression,SVR)模型的输入全变量,分别利用竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)、迭代保留信息变量算法(iteratively retains informative variables,IRIV)和CARS-IRIV联合算法筛选特征变量,建立全变量、CARS、IRIV和CARS-IRIV筛选特征变量与玉米籽粒含水率的SVR模型。引入鲸鱼优化算法(whale optimization algorithm,WOA)优化SVR模型参数c(惩罚因子)和g(核函数参数),结果表明CARS-IRIV筛选特征变量(ε′3.854MHz、ε″3.854MHz、e′5.462MHz、e″5.462MHz、T)建立的SVR模型经WOA优化后(CARS-IRIV-WOA-SVR)具有最优的预测精度,预测集决定系数、预测集均方根误差和剩余预测偏差分别为0.9984,0.40%和24.55,且模型复杂度最低。该研究为基于双介电参数和支持向量回归机实现玉米含水率快速无损检测提供了新的研究思路和基础数据。
To realize the rapid and nondestructive measurement of moisture content in corn,the dielectric constant(ε′)and the dielectric loss(ε″)of corn kernels of different moisture content under different temperature during hot air drying process were measured at an excitation voltage of 500 mV and frequency range of 1 kHz to 5.462 MHz with a precise impedance analyzer and a self-made dielectric parameter measuring sensor,and the mechanism of change in dielectric parameters during drying process was studied and analyzed;meanwhile,a support vector machine-based moisture content prediction model was established by incorporating double dielectric parameters,and a model verification experiment was carried out.It was shown by the experiment results that the dielectric constant(ε′)of the corn kernels in the different moisture content decreased with the increase of measurement frequency;the dielectric loss(ε″)of the corn kernels in the low moisture content increased first and then decreased with the increase of frequency,with an obvious relaxed peak observed within the frequency range of 1 kHz-1 MHz,while for the corn kernels in the high moisture content,the dielectric loss(ε″)decreased with the increase of measurement frequency;in addition,with the increase of frequency,the tangent value of the dielectric loss angle increased first and then decreased for corn kernels in the low moisture content and decreased for the corn kernels in the high moisture content.At the measurement frequency of 5.462 MHz,the dielectric loss angle of the corn kernels decreased with the decrease of the wet-basis moisture content,the first derivative of the dielectric loss angle versus the moisture content dtanδ/dM showed the obvious trend of the segmentation with the decrease of moisture content.Through the analysis of the double dielectric parameter spectrum,the frequency bands for modelling had been selected initially.A total of 31 dimensional variables,including the double dielectric parameters of 15 measured frequency points between 1.072 and 5.462 MHz and the temperature value T,were used as the input full variables of the support vector regression machine(SVR)model.Competitive adaptive reweighting algorithm(CARS),iterative retention information variable algorithm(IRIV)and CARS-IRIV joint algorithm were used to screen the feature variables,respectively.SVR models were established respectively with full variables and the feature variables screened by CARS,IRIV and CARS-IRIV.By optimizing the parameter c(penalty factor)and the parameter g(kernel function parameter)of the SVR model with the WOA(whale optimization algorithm),the result indicated that the optimal prediction performance could be witnessed in the CARS-IRIV-WOA-SVR model.The feature variables screened by CARS-IRIV were(ε′3.854MHz、ε″3.854MHz、e′5.462MHz、e″5.462MHz and T.Determination coefficients of the prediction set,RP 2,RMSEP(root mean square error of prediction)and the RPD(residual predictive deviation)were 0.9984,0.40%and 24.55,respectively.This study provided a new research idea and basic data for rapid nondestructive testing of maize moisture content based on double dielectric parameters and SVR,and it also laid a foundation for the development of intelligent grain moisture content rapid nondestructive testing instrument.
作者
宋华鲁
闫银发
宋占华
孙君亮
李玉道
李法德
Song Hualu;Yan Yinfa;Song Zhanhua;Sun Junliang;Li Yudao;Li Fade(College of Mechanical and Electronic Engineering,Shandong Agricultural University,Tai’an 271018,China;Shandong Provincial Key Laboratory of Horticultural Machineries and Equipments,Tai’an 271018,China)
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2019年第20期262-272,共11页
Transactions of the Chinese Society of Agricultural Engineering
基金
山东省农机装备研发创新计划项目(2015TH101)
关键词
无损检测
算法
水分
双介电参数
竞争性自适应重加权算法
迭代保留信息变量算法
鲸鱼优化算法
nondestructive testing
algorithm
moisture
double dielectric parameters
competitive adaptive reweighted sampling
iteratively retains informative variables
whale optimization algorithm