近红外技术结合SaE-ELM用于烤烟烘烤关键参数的在线监测

摘要

Self-adaptive evolutionary extreme learning machine (SaE-ELM) is a type of single hidden layer feedforward neural network learning algorithm with adaptive differential evolution algorithm to optimize the hidden node parameters. Monitoring key parameters during bulk flue-curing of tobacco is difficult, therefore the dynamic variations of leaf moisture content, chlorophyll (SPAD) and starch in tobacco leaves were monitored by combining near infrared spectroscopy with SaE-ELM and adopting cross validation to select the number of hidden layer nodes. The results showed that comparing with partial least squares (PLS) regression, BP neural network, support vector machine (SVM) regression and extreme learning machine (ELM) quantitative models, SaE-ELM models for moisture, chlorophyll and starch contents had the advantages of automatic parameter optimization, better performance, stronger generalization ability and accurate prediction results, achieving correlation coefficients of 0.931 2, 0.917 6 and 0.916 7, respectively. Near infrared technology combined with SaE-ELM can thus accurately determine the variations of these key parameters during bulk flue-curing of tobacco, which provides technical reference for controlling tobacco bulk curing.%自适应进化极限学习机（SaE-ELM）是一种利用自适应差分进化算法优化隐层输入参数的单隐层前馈神经网络学习算法。为了解决烟叶密集烘烤过程中关键参数难以测定的难题，应用近红外光谱技术结合SaE-ELM，采用交叉验证选择隐含层节点个数，对烘烤过程中含水率，以及叶绿素和淀粉含量3个关键参数的动态变化进行了监测。结果表明：烟叶含水率、叶绿素和淀粉模型预测相关系数分别为0.9312、0.9176和0.9167，与偏最小二乘（PLS）回归、BP神经网络、支持向量机（SVM）回归和极限学习机（ELM）模型相比，SaE-ELM模型参数自动优化、性能优越、泛化能力强、预测结果最好。因此，采用近红外技术结合SaE-ELM能准确测定烟叶烘烤过程中关键参数的变化规律，可为烟叶烘烤调控工艺提供技术参考。