Research on CNN-LSTM-Attention based surface temperature prediction model for fermented grains

Abstract

The prediction of the temperature on the surface of the fermented grains during the steaming process holds significant importance in improving the alcohol yield and quality. To accurately forecast the temperature variations on the surface of the fermented grains, a predictive model based on CNN-LSTM-Attention is proposed. Leveraging convolutional neural networks, latent features of the temperature data on the fermented grains surface are extracted. These extracted feature vectors, representing a time series, are then inputted into a long short-term memory network to further capture the temporal characteristics of the sequence. Finally, employing an attention mechanism, the influential features are highlighted to achieve precise prediction of the fermented grains surface temperature. Historical temperature data from the steaming process of the fermented grains is utilized, and experimental comparisons are conducted with other neural network prediction models. The results demonstrate that the CNN-LSTM-Attention model achieves optimal root mean square error of 1.282 and average absolute error of 0.647, surpassing other models. The experimental findings substantiate the superior accuracy of the CNN-LSTM-Attention model in forecasting the changing trend of the fermented grains surface temperature.