Related Experiment Videos
XLSTM transformer based quality prediction for tobacco cut rag in intermittent processing
Peipei Li1, Peitong Sun2, Mingrui Zhu3
1China Tobacco Henan Industrial CO.Ltd, Henan, 455000, China.
Scientific Reports
|July 10, 2026
Summary
This study introduces an xLSTM-Transformer model for cigarette production quality prediction. The model enhances prediction accuracy, offering a novel approach for real-time monitoring in batch processes.
Area of Science:
- Industrial Engineering
- Artificial Intelligence
- Time Series Analysis
Background:
- Quality prediction in cigarette manufacturing presents significant challenges.
- Existing time-series forecasting models struggle with complex process dynamics.
Purpose of the Study:
- To develop an advanced quality prediction method for cigarette processing.
- To improve the accuracy and efficiency of online quality monitoring.
Main Methods:
- Constructed an xLSTM-based feature correlation layer for local temporal dynamics.
- Designed a Transformer-based feature extraction layer for global interactions.
- Implemented a gated fusion module and a lightweight decoder for efficient prediction.
Main Results:
- Achieved Root Mean Square Error (RMSE) of 0.0165 and Mean Absolute Error (MAE) of 0.0126.
- Demonstrated significant accuracy improvements over iTransformer and RobustTS.
- Maintained a stable R² coefficient above 0.95.
Conclusions:
- The xLSTM-Transformer model effectively models local and global features for enhanced quality prediction.
- This method offers a novel technical approach for dynamic dimension matching and multi-scale feature fusion.
- Provides a robust solution for online quality monitoring in batch production processes.