基于半监督深度网络的冷连轧轧制力预报  

Prediction of cold continuous rolling force based on semi-supervised deep network

作　　者：魏立新[1] 翟博豪 赵志伟[2] 刘建朋 孙浩[1] WEI Li-xin;ZHAI Bo-hao;ZHAO Zhi-wei;LIU Jian-peng;SUN Hao(Intelligent Control System and Intelligent Equipment Engineering Research Center of Ministry of Education,Yanshan University,Qinhuangdao 066004,China;Department of Computer Science and Technology,Tangshan College,Tangshan 063000,China;Tang Steel International Engineering Technology Corp.,Tangshan 063000,China)

机构地区：[1]燕山大学,智能控制系统与智能装备教育部工程研究中心,河北秦皇岛066004 [2]唐山学院计算机科学与技术系,河北唐山063000 [3]唐钢国际工程技术股份有限公司,河北唐山063000

出　　处：《塑性工程学报》2020年第11期70-76,共7页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(61803327);河北省自然科学基金青年基金资助项目(E2018203162);河北省自然科学基金—钢铁联合研究基金资助项目(E2019105123);河北省高等学校科学技术研究项目(ZD2019311)。

摘　　要：针对冷连轧生产中难以建立准确的轧制力数学模型的问题,提出了基于半监督深度网络的轧制力预报模型。首先,使用堆叠去噪自编码器逐层提取输入数据的高阶特征表示。为提高特征提取的有效性,根据输入值与目标值的相关性程度,对其各维度特征损失函数施加不同比例,构成比例损失堆叠去噪自编码器。然后,使用比例损失堆叠去噪自编码器提取的高阶特征初始化深度网络,对目标值进行预测。仿真结果表明,该模型预测精度可控制在3%以内,实现了轧制力的高精度预测。In order to solve the problem of establishing mathematical model of rolling force accurately,a rolling force prediction model based on semi-supervised deep network was proposed. Firstly,a high level feature representation of input data was extracted layer by layer using a stacked denoise autoencoder. To improve the validity of feature extraction,according to the degree of correlation between the input value and the target value,different proportions were applied to the feature loss function of each dimension to form a proportional loss stack denoise autoencoder. Then,the deep network was initialized using the features extracted by proportional loss stack denoise autoencoder to predict the target value. The simulation results show that the prediction accuracy of the model can be controlled within 3%,and the high precision prediction of rolling force is realized.

关 键 词：冷连轧 轧制力预测 半监督学习 深度网络 比例损失堆叠去噪自编码器 

分 类 号：TG335.13[金属学及工艺—金属压力加工]