基于CNN-LSTM混合神经网络的卷烟滤棒质量预测  

Cigarette Filter Rod Quality Prediction Based on CNN-LSTM Hybrid Neural Network

作　　者：王红斌[1] 李志文 WANG Hongbin;LI Zhiwen(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500

出　　处：《长春理工大学学报（自然科学版）》2024年第4期67-73,共7页Journal of Changchun University of Science and Technology(Natural Science Edition)

摘　　要：为了在卷烟生产数据中挖掘出滤棒生产过程的有效信息,解决各批次产品质量检测困难的问题,结合卷积神经网络的特征提取能力与长短期记忆网络处理时序数据的有效性,提出了基于CNN-LSTM混合神经网络的卷烟滤棒质量预测模型。该模型通过卷积层提取输入数据的局部特征,然后在LSTM层中捕捉特征之间的时序关系,分层结构使其具有同时处理不同时间维度信息的能力,从而提升了预测精度。将滤棒的质量定义为圆周值与吸阻值两个物理量,利用卷烟厂6万余条实时生产数据进行模型训练和预测,结果表明:以平均绝对百分误差(MAPE)作为评价标准,圆周指标预测误差为0.078%,吸阻指标预测误差为1.42%,对比各类传统机器学习方法,CNN-LSTM混合神经网络表现出了更高的精确性。该方法可为快速准确地预测卷烟滤棒质量提供技术支持,提升烟草工业的自动化水平。In order to mine effective information about the filter rod production process from cigarette production data.Combining the feature extraction capabilities of convolutional neural networks and the effectiveness of long short-term memory networks in processing time series data,a cigarette filter rod quality prediction model based on CNN-LSTM hybrid neural networks was proposed.The model extracts local features of input data through convolutional layers,and then captures the temporal relationship between features in the LSTM layer.The hierarchical structure gives it the ability to process information in different time dimensions simultaneously.The quality of the filter rod is defined as two physical quantities:circumference value and suction resistance value.More than 60 000 pieces of real-time production data from the cigarette factory are used for model training and prediction.The results show that:using the mean absolute percentage error(MAPE) as the evaluation standard,the circumference indicator prediction error is 0.078%,and the resistance indicator prediction error is1.42%.Compared with various traditional machine learning methods,the CNN-LSTM hybrid neural network shows higher accuracy.This method can provide technical support for quickly and accurately predicting the quality of cigarette filter rods and improve the automation level of the tobacco industry.

关 键 词：滤棒成型 深度学习 卷积神经网络 长短期记忆网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]