Cigarette Filter Rod Quality Prediction Based on CNN-LSTM Hybrid Neural Network
In order to mine effective information about the filter rod production process from cigarette production data. Com-bining 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 informa-tion in different time dimensions simultaneously. The quality of the filter rod is defined as two physical quantities:circum-ference value and suction resistance value. More than 60000 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 is 1.42％. 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.
filter rod formingdeep learningconvolutional neural networklong short-term memory network
NETL
NSTL
万方数据
