Yaojing Yang

TL;DR: A double-blockchain structure based on “public blockchain + consortium blockchain” can not only improve the security of querying tobacco traceability information but also not affect the efficiency of queries.

TL;DR: In this article , a deformable attention mechanism is introduced into the transformer, which enables the model to use high-resolution image features and multi-scale feature information, thereby improving the model detection performance.

TL;DR: Through in-depth customization of the unified query system, based on data lake technology and federated learning technology, the enterprise's multi-heterogeneous data system is re-planned to run stably and reliably and manage enterprise data more efficiently.

Abstract: Abstract The tobacco drying process in the cigarette production has an important effect on the final product quality. Therefore, the intelligent control methods have been widely investigated to ensure the stability of tobacco’s outlet moisture content. The existing work mostly uses a relatively fixed set point of the outlet moisture content for different tobacco batches, which can lead to unforeseen product quality after several processes following the drying process and inaccessible amount of dehydration for the rotary dryer. Some tobacco moisture prediction methods have been studied recently while the relationship with the intelligent control methods remain largely unexplored. To deal with these issues, a novel method is proposed in this paper to identify the optimal set point of the drying outlet moisture content for each tobacco batch. An encoder-decoder model is first developed to forecast the post-drying moisture trajectory. Then, an adaptive filter with specially designed mechanisms and a confidence interval of the dehydration level are constructed to obtain the design constraints. Based on all above, a constrained optimization problem is formulated and solved by the genetic algorithm. Extensive experiments on 895 tobacco batches from a large cigarette factory are carried out, which involves both algorithmic evaluation and field test. It turns out that the proposed method achieves the superior performance and leads to an improvement of the product quality in real production.