| 27 | 0 | 64 |
| 下载次数 | 被引频次 | 阅读次数 |
针对热轧带钢凸度预测过程中多工艺变量耦合作用引起的特征冗余及预测精度受限问题,提出一种基于最大相关最小冗余-递归特征消除(mRMR-RFE)与多学习器融合的凸度预测方法。首先,采用mRMR-RFE从候选工艺变量中筛选关键特征;随后,构建由支持向量回归、双向长短期记忆网络和Transformer组成的异构基学习器,并通过轻量梯度提升机实现Stacking融合,同时引入随机森林开展残差补偿。结果表明,所提出模型对C40凸度预测的平均绝对误差、均方根误差和决定系数分别达到3.75 μm、5.07 μm和0.810,在综合评价指标上表现出较好的预测性能。进一步基于187组现场生产数据开展工业验证,结果表明模型预测值与实测值具有较好一致性,其中79.68%的样本相对误差控制在±10%以内。研究结果可为热轧带钢凸度在线预测及板形质量控制提供参考。
Abstract:To address the feature redundancy and limited prediction accuracy caused by the coupling effects of multiple process variables in hot-rolled strip crown prediction, a crown prediction method based on maximum relevance minimum redundancy-recursive feature elimination (mRMR-RFE) and multi-learner fusion is proposed. First, mRMR-RFE is employed to select key features from candidate process variables. Subsequently, a heterogeneous base learner system consisting of support vector regression, bidirectional long short-term memory network, and Transformer is constructed, where light gradient boosting machine is adopted to achieve Stacking fusion, and random forest is further introduced for residual compensation. The results show that the proposed model achieves a mean absolute error, root mean square error, and coefficient of determination of 3.75 μm, 5.07 μm, and 0.810, respectively, for C40 crown prediction, demonstrating favorable prediction performance in terms of comprehensive evaluation metrics. Furthermore, industrial validation based on 187 sets of field production data indicates good agreement between the predicted and measured values, with 79.68% of the samples achieving relative errors within ±10%. The results provide a reference for online prediction of hot-rolled strip crown and strip shape quality control.
[1] 程传永, 郝熙娟, 孙丽荣, 等. 基于数据驱动的热连轧带钢厚度时间序列预测[J]. 材料与冶金学报, 2026. (CHENG C Y, HAO X J, SUN L R, et al. Time series prediction of hot rolled strip thickness based on data-driven[J]. Journal of Materials and Metallurgy, 2026.)
[2] 安开场, 高卫军, 韩月娇, 等. 新型Hyper UCM轧机板形调控能力分析[J]. 材料与冶金学报, 2026. (An K C, Gao W J, Han Y J, et al. Analysis of new Hyper UC-mill flatness control capability[J]. Journal of Materials and Metallurgy, 2026.)
[3] 杨熙成, 叶俊成, 谢璐璐, 等. 基于GA-BP神经网络的冷连轧带钢板形预测[J]. 材料与冶金学报, 2025, 24(1): 55-61. (Yang X C, Ye J C, Xie L L, et al. Prediction of steel strip flatness in cold continuous rolling based on GA-BP neural network[J]. Journal of Materials and Metallurgy, 2025, 24(1): 55-61.)
[4] 郝学斌. 深度信念网络算法下的热连轧板凸度预测模型[J]. 材料与冶金学报, 2022, 21(5): 364-370. (Hao X B. Prediction model of crown of hot strip rolling based on deep belief network algorithm[J]. Journal of Materials and Metallurgy, 2022, 21(5): 364-370.)
[5] 李嘉林, 高杰, 丁敬国. 蚁群优化算法协同深度极限学习机的热连轧宽度预测模型[J]. 材料与冶金学报, 2024, 23(5): 497-504. (Li J L, Gao J, Ding J G. Deep extreme learning machine based on ant colony optimization for hot strip mill width prediction model[J]. Journal of Materials and Metallurgy, 2024, 23(5): 497-504.)
[6] 王仁忠, 刘广, 王修亮, 等. 基于SHAP可解释性分析与集成学习的箔轧板形预测模型[J]. 材料与冶金学报, 2026. (Wang R Z, Liu G, Wang X L, et al. A foil rolling flatness prediction model based on SHAP interpretability analysis and ensemble learning[J]. Journal of Materials and Metallurgy, 2026.)
[7] Li J D, Wang H M, Shao J, et al. Interpretable data-driven framework with multi-scale residual correction and SHAP analysis for strip profile prediction in hot rolling steel industry[J]. Journal of Manufacturing Systems, 2026, 86: 472-491.
[8] Li J D, Wang X C, Yang Q, et al. IoT-based framework for digital twins in steel production: A case study of key parameter prediction and optimization for CSR[J]. Expert Systems with Applications, 2024, 250: 123909.
[9] 石峰, 程传永, 武文腾, 等. 基于多任务学习的热连轧带钢质量预测建模[J]. 轧钢, 1-13, 2026. (SHI F, CHENG C Y, WU W T, et al. Modeling for quality prediction of hot-rolled strip based on multi-task learning[J]. Steel Rolling, 1-13, 2026.)
[10] 陈明方, 宫宇, 徐向君, 等. LIBS结合最大相关最小冗余特征选择的镁合金快速分类识别[J]. 光学精密工程, 2026, 34(4): 548-558. (CHEN M F, GONG Y, XU X J, et al. Feature selection enhances classification accuracy of magnesium alloys in LIBS spectra[J]. Optics and Precision Engineering, 2026, 34(4): 548-558.)
[11] 李福民, 杨柳, 刘小杰, 等. 基于递归特征消除-梯度提升决策树的高炉悬料智能诊断模型[J]. 冶金自动化, 2024, 48(4): 90-100. (LI F M, YANG L, LIU X J, et al. Intelligent diagnosis model of BF hanging based on RFE-GBDT[J]. Metallurgical Industry Automation, 2024, 48(4): 90-100.)
[12] Wu Z D, Wang X C, Yang Q, et al. Deformation resistance prediction of tandem cold rolling based on grey wolf optimization and support vector regression[J]. Journal of Iron and Steel Research International, 2023, 30(9): 1803-1820.
[13] Wang Z, Zhang J, Lu Q, et al. CNN-BiLSTM-attention based synchronous prediction model for multi-stand rolling force and deformation resistance in cold rolling[J]. Journal of Intelligent Manufacturing, 2025.
[14] Li J, Sun Y, Wang X, et al. Application of novel interpretable machine learning framework for strip flatness prediction during tandem cold rolling[J]. Measurement, 2025, 244: 116516.
[15] Wang Q L, Sun J, Hu Y J, et al. Deep learning-based flatness prediction via multivariate industrial data for steel strip during tandem cold rolling[J]. Expert Systems with Applications, 2024, 237: 121777.
基本信息:
DOI:10.14186/j.cnki.1671-6620.2026.216
中图分类号:TG335.56;TP18
引用信息:
[1]孙亚敏,孙友昭,王晓晨,等.基于mRMR-RFE特征选择与多学习器融合的热轧带钢凸度预测[J].材料与冶金学报().DOI:10.14186/j.cnki.1671-6620.2026.216.
基金信息:
国家重点研发计划资助项目(2023YFB3712404)
2026-10-30
2026-10-30
2026-10-30