基于MIV-SVM的烤烟评吸质量预测模型  

MIV-SVM-based prediction model for smoking quality of flue-cured tobacco

作　　者：聂铭[1] 周冀衡[1] 杨荣生[2] 夏开宝[2] 李强[1] 张卓[1] 

机构地区：[1]湖南农业大学烟草科学与健康重点实验室,长沙410128 [2]云南省烟草公司曲靖市公司,曲靖655000

出　　处：《中国烟草学报》2014年第6期56-62,共7页Acta Tabacaria Sinica

基　　金：烟草工商研共建原料与品牌协同发展机制(ESTB)项目(曲靖烟草公司)(110200801036);曲靖市"天香"特色优质烟叶研究与开发(09YN025)

摘　　要：在matlab环境下利用libsvm软件包建立利用烤烟化学指标对烤烟评吸质量预测的支持向量机模型,利用MIV法筛选变量,选取挥发性酸、石油醚提取物、钾、还原糖、烟碱为模型输入变量;参数优化后最佳c值为0.10882,g值为9.1896;训练样本预测值与实际值的相关系数R为0.9903,测试样本预测值与实际值的相关系数R为0.9966;预测误差分布在[-1,1]。以上结果说明预测模型是可靠的,能够为烤烟评吸质量的评价提供参考。Support vector machine （SVM） which was used to predict smoking quality based on chemical index of flue-cured tobacco was established by using libsvm software package in Matlab environment. Volatile acid, extractable petroleum ether, potassium, reducing sugar and nicotinamide were taken as input variables by using MIV algorithm. Optimal C value was 0.10882, g value was 9.1896 and correlation coefficient R was 0.885 for training samples and 0.8791 for testing samples. Predicted error was distributed in the [-4,4]. Results showed that the prediction model is reliable and capable of providing reference for smoking quality evaluation.

关 键 词：评吸质量 支持向量机 预测模型 参数优化 

分 类 号：TS411[农业科学—烟草工业]