基于数据驱动的松散回潮出口水分预测模型分析

Analysis of Data-driven Based Prediction Models for Moisture Content at the Loosening and Regaining Outlet

为了改善松散回潮工序出口含水率控制精度低、水分波动大对香烟的生产带来的不利影响,采用Pearson相关性分析法筛选出松散回潮机出口含水率的主要影响因素,并将其作为模型的输入,然后基于现场生产数据驱动分别采用主元回归分析法、主元神经网络法以及BP神经网络法建立烟叶出口含水率的预测模型。基于某卷烟厂某牌号的生产数据对3种预测模型的预测控制效果进行检验,结果表明:3种模型含水率预测结果与实际值的平均绝对误差均在0.2%以内,主元神经网络预测模型输出结果与真实值的平均相对误差为0.81%,优于主元回归和BP神经网络预测模型。研究结果对于提高松散回潮工序烟叶出口含水率的控制精度有一定的指导意义。

In order to improve the adverse impact of low control accuracy and large fluctuation of moisture content at the outlet of the moisture loosening and regaining process on cigarette production,Pearson correlation analysis was used to screen out the main influencing factors of moisture content at the outlet of the moisture loosening and regaining machine,and the variables were used as input to the model.Based on field production data,principal component regression analysis,principal component neural network method and the BP neural network method were used to establish a prediction model for the moisture content of tobacco leaves at the outlet.Based on the production data of a certain brand in a cigarette factory,the predictive control effects of the three prediction models were tested.The results showed that the mean absolute errors between the predicted water content of the three models and the actual value were all within 0.2%,and the mean relative error between the output results of the principal component neural network prediction model and the actual value was 0.81%,which was superior to the principal component regression and BP neural network prediction models.The results have certain guiding significance for improving the control accuracy of tobacco leaf outlet moisture content in the moisture loosening and regaining process.