基于生产大数据的薄带钢平整机轧制力设定计算模型  

Roll force setting calculation models of temper rolling mill for thin strip steel based on big data of production

作　　者：王晓东 WANG Xiao-dong(Technical Center of Shougang Jingtang United Iron and Steel Co.,Ltd.,Tangshan 063200,China)

机构地区：[1]首钢京唐钢铁联合有限责任公司技术中心,河北唐山063200

出　　处：《塑性工程学报》2022年第8期209-221,共13页Journal of Plasticity Engineering

摘　　要：基于镀锌线平整机生产大数据,进行了神经网络模型、回归分析模型、网络模型与回归分析模型相结合的混合算法模型的平整机轧制力预测模型的分析与比较。结果表明,与BP神经网络模型和基于径向基函数RBF神经网络模型相比,广义回归GRNN神经网络模型最优,预测结果相对偏差的标准差在12%左右。神经元网络与数学模型结合的混合算法模型的预测精度比单纯网络模型或回归模型的低。多元线性回归模型优于偏最小二乘回归法模型,除少数钢种外,其预测结果相对偏差的标准差在13%以下,且易于实现,适用性强。将多元线性回归参数的数学模型在平整机上进行了初步应用,结果表明该模型具有一定的应用前景。Based on the big data of production of the temper rolling mill galvanized line,the prediction models of the neural network model,regression analysis model and hybrid algorithm model combining network model and regression analysis model for rolling force of temper rolling mill were analyzed and compared.The results show that compared with BP neural network model and RBF neural network model based on radial basis function,the generalized regression GRNN neural network model is the best,and the standard deviation of relative deviation of the prediction results is about 12%.The prediction accuracy of hybrid algorithm model combining neural network and mathematical model is lower than that of pure network model or regression model.The multivariate linear regression model is superior to partial least square regression model.Except for a few steel grades,its standard deviation of relative deviation of the prediction results is less than 13%,and it is easy to implement and the applicability is strong.The mathematical model of multivariate linear regression parameters was initially applied to the temper rolling mill.The results show that the model has a certain application prospect.

关 键 词：薄带钢 平整机 轧制力 神经元网络模型 回归分析模型 混合算法 

分 类 号：TG331[金属学及工艺—金属压力加工]