| 引用本文: | 张凯,王亚礼,张晓雯,彭开香.基于特征融合的粗轧出口温度建模方法与应用[J].控制理论与应用,2026,43(4):853~864.[点击复制] |
| ZHANG Kai,WANG Ya-li,ZHANG Xiao-wen,PENG Kai-xiang.A feature fusion-based rough rolling exit temperature modeling method and its applications[J].Control Theory & Applications,2026,43(4):853~864.[点击复制] |
|
| 基于特征融合的粗轧出口温度建模方法与应用 |
| A feature fusion-based rough rolling exit temperature modeling method and its applications |
| 摘要点击 147 全文点击 22 投稿时间：2024-05-03 修订日期：2025-05-27 |
| 查看全文 查看/发表评论 下载PDF阅读器 HTML |
| DOI编号 10.7641/CTA.2024.40252 |
| 2026,43(4):853-864 |
| 中文关键词 热连轧 温度预测 特征融合 自适应更新 云边端 |
| 英文关键词 hot strip rolling mill temperature prediction feature fusion adaptive updation cloud-edge-end |
| 基金项目 国家重点研发计划项目(2021YFB3301200), 国家自然科学基金项目(62073032, U21A20483)资助. |
|
| 中文摘要 |
| 在带钢热连轧过程中, 粗轧出口温度受加热炉和粗轧工序共同影响, 其建模精度是提高成品带钢质量的关
键. 现有方法大多基于机理模型或仅将加热炉与粗轧数据融合, 对两个工序的特征提取不充分, 且不利于所提出方
法在两工序的分布式部署. 为此, 本文提出了一种基于特征融合的热连轧过程粗轧出口温度建模方法. 该方法首先
针对两个工序数据的特点构建了联合长短期记忆 (LSTM) 网络和时间卷积网络 (TCN) 的特征提取模型, 并利用 LSTM实现了两部分特征的融合, 进而利用全连接模块构建粗轧出口温度的预测模型; 其次, 考虑到热连轧过程带钢
品种和规格的多样性, 构建了预测模型的更新机制; 最后, 将所提出预测方法分布式部署到云边端协同的热连轧过
程原型系统中, 利用某钢铁公司2150产线的实际数据验证了方法的有效性. 验证结果表明, 本文方法与传统方法相
比预测精度提高35.69%, 且在采样周期100 ms下, 预测与更新机制满足实时性要求. |
| 英文摘要 |
| In the hot strip rolling mill (HSRM) process, the rough rolling exit temperature is affected by the heating
furnace and the rough rolling process, and accuracy of the temperature model plays the key role in improving the quality
of the steel products. Most of the existing methods are based on mechanism models or only fuse the heating furnace and
rough rolling data, which is insufficient to extract the features of the two processes, and also these models are not easy to
be deployed in real applications. To this end, a feature fusion-based method is proposed for modeling the rough rolling
exit temperature in the HSRM process. Firstly, two feature extraction methods that using long short term memory (LSTM)
and temporal convolutional network (TCN) are developed respectively by considering the characteristics of process data,
and then a separate LSTM is used to fuse the obtained features, also a fully connected module is finally constructed to
develop the prediction model. Furthermore, considering the differences between different types of steel products, an update
mechanism for the developed prediction model is proposed. In the end, the applicability of proposed model is examined
using a cloud-edge-end collaborative prototype system, where actual HSRM data from the 2150 production line of a steel
company are used. The verification results show that the proposed method improves the prediction accuracy by 35.69%
compared with traditional methods. At the sampling interval of 100 ms, the prediction and update mechanism meets the
real-time requirements. |
|
|
|
|
|