摘要
中国钢铁产业在国际竞争中正面临提升智能制造水平的重大挑战,其中烧结终点智能控制是提升钢铁产量、优化产品质量及实现智能制造的关键工艺环节。总结了国内外关于烧结终点监测的研究进展,包括过程监测、产品质量监测及过程优化控制,并分析了基于多模态大模型的烧结终点优化研究的发展动态和关键问题。基于多模态大模型(DeepSeek)架构,融合终点位置、风箱负压、机尾图像等多元异构参数,构建了一种高精度、强鲁棒性的烧结终点状态软测量模型,并结合烧结过程仿真模拟技术,实现了复杂工况下烧结台车底部温度和压力等数据的精确计算,有效预测了烧结终点的状态。针对烧结过程中少标签、时滞数据的特征,设计一种基于迁移学习和案例推理的烧结矿质量在线监测模型,以实现对烧结矿化学成分、粒度分布等关键质量指标的实时预测和监控。在此基础上,又提出一种多参数协同调控的烧结终点优化控制模型,并结合改进的多目标遗传算法(AMOGA)与滚动时域优化策略,实现不同工况条件下的烧结终点动态优化控制。从冶金行业需求出发,该研究为烧结过程的智能化和精细化控制提供了重要的理论和方法支撑,对于提升钢铁行业智能制造水平和提高生产效率具有重要的科学价值和应用前景。
China's iron and steel industry is facing major challenges in upgrading its intelligent manufacturing capabilities amid international competition.Among these,intelligent control of the sintering endpoint stands as a key technological link for increasing steel output,optimizing product quality and achieving intelligent manufacturing.It reviewed the progress in sintering endpoint monitoring research both domestically and internationally,encompassing process monitoring,ore quality index monitoring,and process optimization control.It also analyzed the development trends and key issues of sintering endpoints under the framework of multimodal large models.Based on the multimodal large model(DeepSeek)architecture,it integrated heterogeneous parameters such as endpoint position,windbox negative pressure,and tail-end images to construct a high-accuracy and robust soft-sensing model for the sintering endpoint state.Combined with sintering process simulation technology,the model achieved accurate calculation of temperature and pressure data at the bottom of the sintering trolley under complex working conditions,effectively predicting the sintering endpoint status.Addressing the characteristics of sparse labels and time-delayed data in the sintering process,it designed an online monitoring model for sinter quality based on transfer learning and case-based reasoning,enabling real-time prediction and monitoring of key quality indicators such as chemical composition and particle size distribution.Furthermore,a multi-parameter collaborative control model for sintering endpoint optimization was proposed,incorporating an improved adaptive multi-objective genetic algorithm(AMOGA)and rolling horizon optimization strategy to achieve dynamic optimization control of the sintering endpoint under varying working conditions.From the perspective of the metallurgical industry's needs,it provides important theoretical and methodological support for the intelligent and refined control of the sintering process,offering significant scientific value and application prospects for enhancing the intelligent manufacturing level and optimizing production efficiency in the steel industry.
作者
刘颂
金焕
刘然
刘小杰
赵军
李宝生
郝建海
吕庆
LIU Song;JIN Huan;LIU Ran;LIU Xiaojie;ZHAO Jun;LI Baosheng;HAO Jianhai;LÜQing(School of Artificial Intelligence,Tangshan University,Tangshan 063000,Hebei,China;School of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063017,Hebei,China;Ironmaking Division,Tangshan Iron and Steel Group Co.,Ltd.,Tangshan 063000,Hebei,China;Hebei Dahe Chemical Group Co.,Ltd.,Shijiazhuang 050011,Hebei,China)
出处
《钢铁》
北大核心
2025年第9期34-47,共14页
Iron and Steel
基金
河北省自然科学基金资助项目(E2024105036,E2024209101)
河北省燕赵黄金台聚才计划骨干人才资助项目(B2024005019)
河北省创新能力提升计划资助项目(23560301D)
唐山市人才资助项目(B202302007)。
关键词
烧结终点控制
多模态大模型
多元异构
软测量模型
仿真模拟技术
在线监测模型
多目标遗传算法
滚动时域优化策略
sintering end-point control
multimodal large model
multi-source heterogeneous data
soft-sensing model
simulation technology
online monitoring model
multi-objective genetic algorithm
receding horizon optimization strategy
