摘要
为提高制丝工艺质量评价中参数赋权分析的科学性和客观性,选取"云烟"某规格一类卷烟制丝过程全批次数据的稳态数据样本,通过Pearson相关性矩阵筛选各工序出口含水率的解释变量,然后利用随机森林回归进行建模分析,采用拟合优度和五折交叉验证的测试集标准化均方误差分别验证模型的拟合效果和外推预测性能,最终根据OOB均方误差的平均递减值进行解释变量影响权重的测度和关键参数的筛选。结果表明:1综合Pearson相关性矩阵和设备控制原理,筛选得到37个解释变量;2制丝过程5个工序随机森林回归模型的拟合优度均大于0.9、五折交叉验证测试集的标准化均方误差均小于1,表明模型的拟合效果和外推预测性能较好;3根据解释变量影响权重的测度分析,筛选得到18个关键参数;4基于全样本数据建立的制丝过程关键参数筛选和赋权方法,可为制丝关键质量特性精准控制和工艺质量评价提供参考。
In order to weight the parameters in primary processing scientifically and objectively,the explanatory variables of moisture content in output tobacco from each processing step were screened by Pearson correlation matrix based on the steady state data samples of a whole processed batch of a selected specification of brand＂Yunyan＂.The prediction models were established by random forest regression,and further verified by goodness of fit and NMSE（Normalized Mean Square Error） of test set of 5-fold cross-validation.Finally,the weights of explanatory variables were calculated and key parameters were screened by the average decrement of OOB（Out of bag） mean square error.The results showed that：1） Thirty-seven explanatory variables were screened out by Pearson correlation matrix combined with equipment control principle.2）The goodness of fit of models for five processing steps was above 0.9 and NMSE of test set of 5-fold cross-validation was less than1,which proved the fitting and extrapolation prediction performance of the models.3）Eighteen key parameters were screened out on the basis of the weight of explanatory variables.4） The proposed method for screening and weighting key parameters in primary processing provided a reference for the precise control of primary processing and the assessment of processing quality.
出处
《烟草科技》
EI
CAS
CSCD
北大核心
2017年第2期63-71,共9页
Tobacco Science & Technology
基金
红云红河烟草(集团)有限责任公司科技项目"基于云平台的工艺质量智能管控研究及应用"(HYHH2016GY04)
