摘要
为提高带钢精轧过程宽度的控制精度,以实际生产数据为驱动,建立深度学习网络模型,对自由宽展进行预测。采用拉伊达准则对实际生产数据进行清洗,对清洗后的数据进行相关性分析,并提取相关系数大于给定阈值的特征。基于预处理后的特征数据,对深度学习网络进行训练,建立自由宽展预测模型。针对测试实例,分别采用该模型与传统数学模型进行预测,并从均方误差、最大偏差以及误差分布等多个方面进行对比分析。结果表明,所建立的深度学习预测模型,具有更高的预测精度和更好的性能指标。
In order to improve the control accuracy of width for hot strip finishing mill, deep learning network model was established to predict the lateral spread driven by the actual production data. The actual process data were cleaned by the Laida criterion. The correlation analysis was carried out on the cleaned data and the features with correlation coefficient greater than the given threshold were extracted. Based on the preprocessed feature data, the deep learning network was trained, and then lateral spread prediction model was established. According to the test examples, the model and traditional mathematical model were used for prediction. Comparison and analysis were carried out from the aspects of mean squared error, maximum deviation and error distribution. The results shows that the established deep learning prediction model has higher prediction accuracy and better performance index.
作者
张峥
仲兆准
李阳
章顺虎
ZHANG Zheng;ZHONG Zhao-zhun;LI Yang;ZHANG Shun-hu(School of Iron and Steel,Soochow University,Suzhou 215100,Jiangsu,China)
出处
《中国冶金》
CAS
北大核心
2022年第11期121-127,共7页
China Metallurgy
基金
国家自然科学基金资助项目(52074187)。
关键词
深度学习
带钢精轧
自由宽展
数据清洗
相关性分析
预测模型
deep learning
hot strip finishing mill
lateral spread
data cleaning
correlation analysis
prediction model
