摘要
针对梗丝回潮出口含水率受外界因素影响较大,难以确定加工过程中的加水总量,进而会对后续梗丝加料出口含水率稳定性及梗丝质量产生影响的问题,提出基于卷积神经网络(CNN)和长短期记忆(LSTM)网络的深度集成,并结合注意力机制开发梗丝回潮加水量预测模型。首先,通过结合CNN和注意力机制解决影响梗丝回潮加水量的多变量和强耦合特性对预测建模带来的挑战。其次,通过LSTM网络门控单元处理序列数据,有效学习数据的时间特征。试验结果表明,此预测模型在预测梗丝回潮加水量时的决定系数(R^(2))可达0.97,平均绝对误差(MAE)低至5%。实际生产中,在后端工序梗丝加料出口含水率稳定性得到了大幅提高,合格率由72%提高至95%。
Addressing the problem that the moisture content at the outlet of cut stem rehumidification is significantly influenced by external factors,making it difficult to determine the total water addition during processing,which in turn affects the stability of the moisture content at the outlet of subsequent cut stem casing and the quality of the cut stems,a prediction model for water addition in cut stem rehumidification is proposed.This model is developed based on a deep integration of Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)networks,combined with an attention mechanism.Firstly,the challenges posed by the multivariate and strongly coupled characteristics affecting the water addition in cut stem rehumidification for predictive modeling are addressed by integrating CNN with an attention mechanism.Secondly,the gating units of the LSTM network are employed to process sequential data,effectively learning the temporal features of the data.Experimental results demonstrate that the proposed prediction model achieves a coefficient of determination(R^(2))of up to 0.97 and a Mean Absolute Error(MAE)as low as 5%when predicting the water addition in cut stem rehumidification.In actual production,significant improvements have been observed in the stability of the moisture content at the outlet of the cut stem casing in the backend process,with the pass rate increasing from 72% to 95%.
作者
许晓梅
丁筱茜
杨潇谊
汪润
茶青
XU Xiaomei;DING Xiaoqian;YANG Xiaoyi;WANG Run;CHA Qing(Kunming Cigarette Factory,China Tobacco Yunnan Industrial Co.,Ltd.,Kunming,Yunnan 650000,China)
出处
《自动化应用》
2025年第16期26-30,共5页
Automation Application
