Thermal-aware multi-task soft sensor with dynamic weighting for real-time prediction of temperature and CO in pulverised-coal preheating

Read full text Summarise with VIRAT
Authors: Dongge Feng, Hong Tang, Xueqiang Shen, Zhenbang Wang, Xiaoju Han, Guojia Yao, Lingfang Sun, Guotao Liu
Cited by: 0
References: 41

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate, real-time monitoring of preheating temperature and CO volume fraction is crucial for the stable operation of pulverised-coal preheating systems. However, conventional hardware sensors are prone to failure under harsh conditions such as high temperature and heavy dust. Meanwhile, existing data-driven soft sensors typically model temperature and CO separately, failing to capture their correlation, which to some extent limits prediction accuracy. To address this issue, this study proposes a thermal-aware multi-task soft sensor that performs joint prediction by exploiting the temperature–CO correlation and uses dynamic weighting to suppress cross-task error propagation. The framework employs a temporal convolutional network to learn shared temporal representations and gated recurrent units for task-specific prediction. A thermal-aware mechanism feeds the predicted temperature into the CO branch to enhance cross-task interaction, whilst a dynamic weighting module adaptively regulates the transferred temperature information to alleviate error propagation. To improve efficiency, a Pearson–SHAP-based method is further introduced to select six key input variables. Evaluation on operational data from an industrial preheating unit shows mean RMSEs of 8.97 °C for temperature and 0.40 vol% for CO, consistently outperforming baseline methods. Further online prediction evaluation demonstrates stable real-time prediction, with errors remaining within acceptable engineering tolerance, and the learnt temperature–CO correlation patterns being broadly consistent with established process understanding.

Fields of Study & Subjects

Fault Detection and Control Systems Advanced Control Systems Optimization Air Quality Monitoring and Forecasting

Publication details

Journal / source
Case Studies in Thermal Engineering
Volume
82
Pages
108014–108014
Publication date
2 April 2026
Language
EN