摘要
板形控制过程具有非线性、遗传性、强耦合等特征,现有的数学模型不能充分描述这些关系;同时,蕴含在钢铁大数据中的价值没有得到充分挖掘,在由“数据”到“信息”的处理过程中存在断层,没有形成对板形控制的良好反馈。为提高板形控制水平,首先设计了一种融合集成经验模态分解和小波变换的轧制生产数据降噪方法,实现高噪声条件下有效信息的准确提取;以此为基础,提出使用正交信号校正改进的偏最小二乘算法,结合生产数据实时优化板形调控功效系数,准确描述了实际数据中板形变化量与执行机构调节变化量的比例关系。通过对轧制过程数据的治理与高精度板形调控功效系数的获取,建立起了与六辊轧机实体系统相互映射的虚拟镜像,实现了板形控制数字孪生模型的构建。之后,以板形解析机理模型和经验知识为先导,分别基于核偏最小二乘法和深度神经网络算法建立了板形预测模型,并在预测效果更好的基于深度神经网络算法的板形预测模型基础上,提出了启发式算法和梯度下降算法相结合的板形设定智能优化方法。最终,通过构建以数字孪生模型为核心、以多目标协调优化为特征的板形控制信息物理系统,成功实现了冷连轧多机架工艺参数的动态优化设定,板形标准差由优化前的3.42 IU降至1.91 IU。
The flatness control process has the characteristics such as nonlinearity,heredity and strong coupling.However,the existing mathematical models cannot fully describe these relationships.Furthermore,the value contained in the massive data on steel were not fully explored and also there were faults in the process of processing from Data to Information,which led to bad feedback infromation for control of the flatness.In order to improve the control level for flatness,the noise-reducing method for rolling production data that syncretizing integrated empirical mode decomposition and wavelet transform was desinged for achieving accurate extraction of effective information under high noise conditions.Based on that,the improved partial Least Squares rolling process parameters was successfully achieved.And therefore the standard deviation for flatness or shape decreased from 3.42 IU to 1.91 IU.
作者
张殿华
魏臻
王军生
宋君
王青龙
孙杰
ZHANG Dianhua;WEI Zhen;WANG Junsheng;SONG Jun;WANG Qinglong;SUN Jie(The State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,Liaoning,China;Ansteel Beijing Research Institute Co.,Ltd.,Beijing 102211,China)
出处
《鞍钢技术》
CAS
2023年第5期1-11,共11页
Angang Technology
基金
国家重点研发计划项目(2022YFB3304800)
国家自然科学基金项目(U21A20117,U21A20475,52074085)。
关键词
板形控制
冷轧
数字孪生
信息物理系统
flatness or shape control
cold rolling
digital twin
information physics system
