ABSTRACT The loose moisture regain link uses water to restore the tobacco leaves from a brittle and hard state to a suitable flexibility and optimal moisture absorption state. It directly affects the quality of tobacco leaves and the manufacturing process of cigarettes. As a typical time series problem, its process data have high time series and complex coupling characteristics, which are influenced by multi‐dimensional characteristics. In order to solve the problem of insufficient mining of inter‐feature relationships, we propose an SLM‐Crossformer algorithm. The two‐stage multi‐head attention mechanism focuses on the cross‐dimensional dependencies between multiple features. It learns the latent associations between variables and mines deep cross‐domain semantic information. And a fusion layer is designed to capture global information and inject a macro perspective into the model, improving the accuracy of process quality prediction. We take the quality of the loose recycling process in a certain cigarette factory as the research object. Experimental and comparative analysis shows that the prediction accuracy of SLM‐Crossformer is significantly higher than other models.
Tingting et al. (Thu,) studied this question.