锻压技术

基于深度学习的宽厚板热轧轧制力预测

英文标题：Prediction on rolling force in hot rolling of wide and thick plate based on deep learning

作者：郭金涛王龙余建波冀秀梅

单位：上海大学

分类号：TG335.5+1

出版年,卷(期):页码：2022,47(7):167-174

摘要：

为了提高宽厚板热轧生产过程控制中轧制力的预测精度，构建了融合SIMS模型的深度学习网络模型，对宽厚板热轧轧制力进行预测研究。利用深度学习框架，构建了一种基于残差连接的深度学习网络模型，并融合SIMS模型计算值，通过误差反向传播计算损失函数的梯度，同时使用Mini-Batch与RMSProp结合的优化算法对权重参数进行更新优化。利用残差连接引入纯线性的信息携带轨道，从而创造一条捷径，将较早的信号重新注入给下游的网络层，使用早停机制、批标准化等策略抑制模型过拟合现象，提高模型的预测精度。基于上述建模方法，针对宽厚板热轧生产线的轧制数据进行了建模实验。结果表明，以相对误差绝对值小于5%在测试集中的占比作为评价指标，相比于传统SIMS模型，融合SIMS模型、基于残差连接的深度学习网络可实现轧制力的高精度预测，该模型的预测精度平均提升了21.72%。

In order to improve the prediction accuracy of rolling force in the hot rolling production process control for wide and thick plate, a deep learning network model integrating SIMS model was constructed to predict the rolling force of wide and thick plate in hot rolling. Then, by using the deep learning framework, a deep learning network model based on residual connection was constructed, which integrated the calculated values of SIMS model, calculated the gradient of loss function through error back propagation, and updated and optimized the weight parameters by using the optimization algorithm combining Mini-Batch and RMSProp. Furthermore, a shortcut was created to inject the earlier signals into the downstream network layers by using the residual connection to introduce a pure linear information carrying track, and the over fitting phenomenon of the model was suppressed by using the early-stopping mechanism and batch normalization and other strategies to improve the prediction accuracy of the model. Based on the above modeling method, the rolling data of wide and thick plate in hot rolling production line was modeled experimentally. The results show that taking the ratio of absolute value for relative error less than 5% in the test set as the evaluation index, compared with the traditional SIMS model, the deep learning network integrating SIMS model based on residual connection can achieve high-precision prediction of rolling force, and the prediction accuracy of the model is improved by an average value of 21.72%.

基金项目：

基于数字化发展的制造业生态构建和路径研究（2022-XY-100）

作者简介：郭金涛(1996-)，男，硕士研究生 E-mail：kimtao@shu.edu.cn 通信作者：余建波（1982-），男，博士，教授 E-mail：jbyu@shu.edu.cn

参考文献：

[1]Liu X, Liu X H, Song M, et al. Theoretical analysis of minimum metal foil thickness achievable by asymmetric rolling with fixed identical roll diameters[J]. Transactions of Nonferrous Metals Society of China, 2016, 26(2): 501-507.

[2]Zuo Y B, Xing F U, Cui J Z, et al. Shear deformation and plate shape control of hot-rolled aluminium alloy thick plate prepared by asymmetric rolling process[J].Transactions of Nonferrous Metals Society of China, 2014, 24(7): 2220-2225.

[3]李海军, 徐建忠,王国栋.热轧带钢精轧过程高精度轧制力预测模型[J].东北大学学报：自然科学版, 2009, 30(5): 669-672.

Li H J, Xu J Z, Wang G D. High-precision rolling force prediction model for hot strip continuous rolling process[J].Journal of Northeastern University:Natural Science,2009,30(5):669-672.

[4]周富强, 曹建国,张杰,等. 基于神经网络的冷连轧机轧制力预报模型[J]. 中南大学学报：自然科学版, 2006, 37(6):1155-1160.

Zhou F Q, Cao J G, Zhang J, et al. Prediction model of rolling force for tandem cold rolling mill based on neural networks and mathematical models[J]. Journal of Central South University: Science and Technology,2006, 37(6):1155-1160.

[5]朱永波, 张飞,张勇军,等. 基于粒子群优化的带钢凸度神经网络预测模型研究[J]. 冶金自动化,2019,43(2):11-15，28.

Zhu Y B, Zhang F, Zhang Y J, et al. Particle swarm optimized neural network for strip crown prediction model research[J]. Metallurgical Industry Automation,2019,43(2):11-15，28.

[6]熊文韬, 谢三山,黄兆飞,等.基于神经网络遗传算法函数寻优与回弹补偿技术的某型汽车前幅拉延成形优化[J].塑性工程学报,2020,27(6):38-45.

Xiong W T, Xie S S, Huang Z F, et al. Optimization of drawing forming for front panel of an automobile based on neural network genetic algorithm function optimization and SCP technology[J].Journal of Plasticity Engineering,2020,27(6):38-45.

[7]窦博. 热连轧轧制力贝叶斯神经网络预测与模型优化[J].金属制品,2017,43(6):42-48.

Dou B. Prediction of rolling force and model optimization with Bayes neural network[J].Metal Products,2017,43(6):42-48.

[8]冀秀梅, 王龙,高克伟,等.极限学习机在中厚板轧制力预报中的应用[J].钢铁研究学报,2020,32(5):393-399.

Ji X M, Wang L, Gao K W, et al. Application of ELM to predict plate rolling force[J]. Journal of Iron and Steel Research, 2020, 32(5): 393-399.

[9]王秀梅, 王国栋.综合神经网络在热连轧机组轧制压力预报中的应用[J].钢铁研究学报,1998, 10(4):72-74.

Wang X M, Wang G D. Application of combination neural network to the prediction of the rolling load in the finishing train of hot strip mill[J]. Journal of Iron and Steel Research, 1998, 10(4):72-74.

[10]吕程, 王国栋,刘相华,等.基于神经网络的热连轧精轧机组轧制力高精度预报[J].钢铁,1998,(3):35-37.

Lyu C, Wang G D, Liu X H, et al. High-precision prediction of rolling load of finishing stands with neural networks[J].Iron & Steel,1998,(3):35-37.

[11]刘元铭, 王振华,王涛,等.热轧带钢出口凸度数据驱动建模及智能化预测分析[J].中国机械工程,2020,31(22):2728-2733.

Liu Y M, Wang Z H, Wang T, et al. Data-driven modeling and intelligent prediction analysis for hot strip outlet crowns[J].China Mechanical Engineering,2020,31(22):2728-2733.

[12]陈丹, 邵健,殷实,等.基于大数据平台的冷连轧轧制力自学习模型优化[J].冶金自动化,2020,44(6):25-29.

Chen D, Shao J, Yin S, et al. Optimization of self-learning model of cold rolling force based on big data platform[J].Metallurgical Industry Automation,2020,44(6):25-29.

[13]曹建国, 江军,赵秋芳,等.基于数据挖掘的宽厚板板凸度控制[J].中南大学学报：自然科学版,2019,50(11):2743-2752.

Cao J G, Jiang J, Zhao Q F, et al. Wide and heavy plate crown control based on data mining[J]. Journal of Central South University：Science and Technology, 2019, 50(11):2743-2752.

[14]刘明华, 张强,刘英华,等.基于机器学习的热轧轧制力预测[J].锻压技术,2021,46(10):233-241.

Liu M H, Zhang Q, Liu Y H, et al. Prediction of rolling force in hot rolling based on machine learning[J]. Forging & Stamping Technology,2021,46(10):233-241.

[15]章顺虎, 姜兴睿,尤凤翔,等.融合工业大数据的热轧厚板轧制力模型研究[J].精密成形工程,2020,12(2):8-14.

Zhang S H, Jiang X R, You F X, et al. Investigation on the model of rolling force by integrating industrial big data[J].Journal of Netshape Forming Engineering,2020,12(2):8-14.

[16]魏立新, 魏新宇,孙浩,等.基于深度网络训练的铝热轧轧制力预报[J].中国有色金属学报,2018,28(10):2070-2076.

Wei L X, Wei X Y, Sun H, et al. Prediction of aluminum hot rolling force based on deep network[J]. The Chinese Journal of Nonferrous Metals,2018,28(10):2070-2076.

[17]马威, 李维刚,赵云涛,等.基于深度学习的热连轧轧制力预测[J].钢铁研究学报,2019,31(9):805-815.

Ma W, Li W G, Zhao Y T, et al. Prediction of hot- rolled roll force based on deep learning[J].Journal of Iron and Steel Research, 2019,31(9):805-815.

[18]马湧, 王晓鹏,马莎莎.基于Keras深度学习框架下BP神经网络的热轧带钢力学性能预测[J].冶金自动化,2019,43(2):6-10.

Ma Y, Wang X P, Ma S S. Prediction of mechanical properties of hot rolled strip based on BP neural network under Keras deep learning framework[J]. Metallurgical Industry Automation, 2019, 43(2):6-10.

[19]王秀梅, 王国栋,刘相华.人工神经网络和数学模型在热连轧机组轧制力预报中的综合应用[J].钢铁,1999，(3):39-41.

Wang X M, Wang G D, Liu X H. Application of neural networks in combination with mathematical models to prediction of rolling load of hot strip rolling mill[J].Iron & Steel,1999，(3):39-41.

[20]Zhou B C, Han C Y, Guo T D. Convergence of stochastic gradient descent in deep neural network[J]. Acta Mathematicae Applicatae Sinica：English Series, 2021, 37(1):126-136.

[21]Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[A]. International Conference on Machine Learning[C]. PMLR, 2015.

[22]He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[A]. IEEE Conference on Computer Vision & Pattern Recognition[C]. IEEE Computer Society, 2016.

[23]贺毓辛. 轧制工程学[M]. 北京：化学工业出版社, 2010.

He Y X. Rolling Engineering[M].Beijing: Chemical Industry Press,2010.

[24]李飞飞, 宋勇,刘超,等.热轧带钢力学性能预报模型的误差分布建模研究[J].冶金自动化,2019,43(6):28-33.

Li F F, Song Y, Liu C, et al. Research on error distribution modeling of mechanical performance prediction model for hot rolled strip[J].Metallurgical Industry Automation,2019,43(6):28-33.

[25]Yarotsky D. Error bounds for approximations with deep ReLU networks[J]. Neural Networks: The Official Journal of the International Neural Network Society, 2017, 94(2): 94-103.

服务与反馈：

【文章下载】【加入收藏】