基于SCAD方法的烘丝段水分控制参数筛选及应用

Selection and Application of Moisture Control Parameters in the Drying Section Based on the SCAD Method

摘要: 为实现对制丝过程烘丝段出料水分的精准控制，科学、客观地识别并筛选影响水分的关键工艺参数。选取以某牌号卷烟的制丝过程全批次稳态数据为研究对象，在传统逐步回归方法的基础上，引入三种通过添加惩罚项来压缩变量系数的Lasso族方法，即：Lasso方法、适应性Lasso方法和SCAD方法，分别构建以烘丝段出料水分为因变量的变量选择模型，并采用AIC、BIC和MSE三种评价指标对模型进行比较，最后依据最优模型进行关键工艺参数的筛选及重要性排序。结果表明：① Lasso族方法在模型拟合优度、预测精度和复杂度控制方面均显著优于传统逐步回归方法，其中SCAD方法的综合性能表现最优；② 基于SCAD方法确定了烘丝段的4个关键工艺参数，按其重要性排序依次为：II区筒壁温度、膨胀单元蒸汽体积流量、切叶丝含水率和工艺气速度。

Abstract: To achieve precise control over the moisture content of the discharge in the drying section of the tobacco primary processing, it is essential to scientifically and objectively identify and screen the key process parameters affecting moisture. This study selects steady state data from the entire batch of a specific brand of cigarette production process as the research object. Building on the traditional stepwise regression method, three Lasso family methods—Lasso, Adaptive Lasso, and SCAD are introduced, which compress variable coefficients by adding penalty terms. Variable selection models are constructed with the moisture content of the drying section discharge as the dependent variable. The models are compared using three evaluation metrics: AIC, BIC, and MSE. Finally, the optimal model is used to screen and rank the importance of key process parameters. The results show that: ① The Lasso-family methods significantly outperform the traditional stepwise regression method in terms of model goodness-of-fit, prediction accuracy, and complexity control, with the SCAD method demonstrating the best overall performance; ② Based on the SCAD method, four key process parameters for the drying section are identified, ranked in order of importance as follows: Zone II wall temperature, expansion unit steam volumetric flow rate, cut tobacco moisture content, and process air velocity.