A novel deep CGRU model with feature attention for prediction of moisture content in Leaf moistening

The moisture content at the outlet of the leaf moistening process is an important process indicator in the production process, and its stability and accuracy directly affect the product quality. To address the issues, this paper proposes a moisture content prediction method based on feature attention mechanism and deep spatio-temporal representation. First, feature attention is applied to dynamically adjust the importance of different process variables based on their relevance to the target quality variable. Convolutional Neural Networks (CNN) and Gated Recurrent Unit (GRU) structure is utilized to capture the coupling and correlation among the variables, and capture the long-term temporal dependencies. Additionally, a parallel GRU is constructed to explore the autocorrelation properties among historical quality variables. Finally, the paper utilizes a concatenation layer to merge the parallel information into a deep spatio-temporal representation, which is then fed into a fully connected layer network to predict the moisture content at the outlet. The experimental results demonstrate that the proposed method outperforms the baseline methods in terms of prediction accuracy and stability, and provides effective decision support for operators and meets the requirements of industrial production.