GAT-LSTM-Based Outlet Moisture Content Prediction Model for Tobacco Production

Loosening and moistening are critical stages in tobacco processing, as the moisture content at this stage significantly affects the quality and flavor of cigarettes. Controlling the outlet moisture during the loosening and moistening process is essential. However, predicting the moisture content accurately during this stage can be challenging. To address this issue, we propose an innovative predictive model that integrates Graph Attention Networks (GAT) with Long Short-Term Memory networks (LSTM). Our approach aims to fully exploit the latent correlations within production data, enhancing both the stability and accuracy of predictions. We constructed a graph structure from temporal production data based on a correlation analysis of production parameters. This methodology leverages the strong spatial feature capture capability of GAT networks, while the LSTM component effectively captures temporal sequence variations and reveals intricate interactions among different production parameters. When applied to real-world production data, our proposed model demonstrated superior performance across various test sets compared to existing machine learning and deep learning models. Specifically, it achieved exceptional results in evaluation metrics including Mean Absolute Error (MAE), Mean Squared Error (MSE), and Coefficient of Determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), with scores of 0.0151, 0.0011, and 0.9932, respectively. These results not only validate the efficacy of our model but also highlight its exceptional capability in managing complex industrial data.