中南大学学报（自然科学版）2025，Vol.56Issue(7)：2719-2730,12.DOI:10.11817/j.issn.1672-7207.2025.07.009
融合机理模型与深度学习的加热炉钢坯温度预测
Hybrid physical-deep learning approach to billet temperature forecasting in reheating furnaces
摘要
Abstract
Data-driven models exhibit limitations in mechanism ambiguity and parameter sensitivity for billet temperature prediction in reheating furnaces,leading to prediction accuracy degradation.To address this,a novel billet temperature prediction algorithm integrating mechanism models with deep learning was proposed.Firstly,this algorithm employs the 1D unsteady Convection-Radiation Heat Transfer model(CRHT)to perform preliminary computations.The results were fused with furnace operating parameters to incorporate mechanistic knowledge.Secondly,the differentiated creative search(DCS)algorithm was enhanced by using tent chaotic mapping and dynamic adaptive weights,and co-optimization of hyperparameters for a fused Bidirectional temporal convolutional network(BITCN)and bidirectional long short-term memory(BILSTM)architecture were achieved.Finally,model accuracy was systematically verified by actual production data.The results show that during billet temperature prediction in the soaking zone of the reheating furnace,the proposed model achieves 52.8%reduction in mean absolute error(MAE)and 28.9%reduction in root mean square error(RMSE)against conventional BITCN-BILSTM,exhibiting significant prediction accuracy improvement.关键词
钢坯温度预测/机理模型/双向时间卷积神经网络/双向长短期记忆/差异创意搜索Key words
steel billet temperature prediction/mechanistic model/bidirectional temporal convolutional networks/bidirectional long short-term memory/differentiated creative search分类
矿业与冶金引用本文复制引用
冯旭刚,杨克,安硕,王正兵,唐得志,王伟,柳传武,潘磊..融合机理模型与深度学习的加热炉钢坯温度预测[J].中南大学学报（自然科学版）,2025,56(7):2719-2730,12.基金项目
安徽省高校自然科学研究重点项目(2023AH051107) （2023AH051107）
芜湖市重点研发与成果转化项目(2023yf017)(Project(2023AH051107)supported by the Key Project of Natural Science Research in Universities of Anhui Province （2023yf017）
Project(2023yf017)supported by the Wuhu City Key Research and Development and Achievement Transformation Project) （2023yf017）
