摘要
对流化床进行工程化改造,安装近红外光谱仪,对制粒过程中的颗粒水分含量进行光谱采集。采用偏最小二乘法(PLS)、粒子群-核岭回归(PSO-KRR)和随机森林-偏最小二乘法(RF-PLS)建立回归模型,对不同的算法的预测精度进行了研究。结果表明,PLS模型预测集均方根误差(RMSE)为0.2180,相关系数R为0.9717;PSO-KRR模型预测集RMSE为0.2154,R为0.9738;RF-PLS模型预测集RMSE为0.2059,R为0.9733。其中RF-PLS模型的预测效果最好。
The fluidized bed was engineered and a near-infrared probe was installed to collect the spectrum of particles in the granulation process.The particle swarm-kernel ridge regression(PSO-KRR)and random forest-partial least squares(RF-PLS)were used to establish regression models with spectral data,and studied the prediction accuracy of different algorithms.The results show that the RMSE of PLS model prediction set is 0.2180 and the correlation coefficient R is 0.9717;the RMSE of PSO-KRR model prediction set is 0.2154 and the correlation coefficient is 0.9738;and the RMSE of RF-PLS model prediction set is 0.2059 and the correlation coefficient is 0.9733.The RF-PLS model has the best prediction effect.
作者
李民东
王海燕
陈庆伟
周军
皇攀凌
LI Min-dong;WANG Hai-yan;CHEN Qing-wei;ZHOU Jun;HUANG Pan-ling(College of Mechanical Engineering,Shandong University,Jinan 250061,China;College of Pharmacy,Shandong University,Jinan 250012,China;Key Laboratory of High Efficiency and Clean MechanicalManufacture,Ministry of Education,Shandong University,Jinan 250061,China)
出处
《应用化工》
CAS
CSCD
北大核心
2020年第5期1325-1328,共4页
Applied Chemical Industry
基金
国家科技重大专项(2018ZX09201010)
山东省重点研发计划项目(2018CXGC1405)
山东省重点研发计划项目(2017CXGC0215)。
关键词
流化床
制粒
近红外光谱仪
水分
偏最小二乘法
粒子群算法
核岭回归
随机森林
fluidized bed
granulation
NIRS
water content
partial least squares method
particle swarm optimization
kernel ridge regression
random forest