基于随机森林的热轧带钢终轧温度预报
Prediction on finish rolling temperature for hot-rolled strip based on random forest
马更生 1樊涛 2马晓宝 3路平 1李炯 1陆伟 1韩东序1
作者信息
- 1. 常州工业职业技术学院机械与交通学院,江苏 常州 213164
- 2. 中国船舶重工集团应急预警与救援装备股份有限公司,湖北 武汉 430223
- 3. 太原理工大学机械工程学院,山西 太原 030600
- 折叠
摘要
复杂的精轧区边界条件和难以观测的参数导致传统在线带钢热终轧温度的预报精度受限.为了提高终轧温度预报精度,采用随机森林进行数据驱动方式的建模.选取影响终轧温度的43个特征因子作为数据驱动终轧温度预测模型的输入值,采用NCL和SMOTE混合算法处理换规格等情况的非平衡数据集,决策树的随机特征选取包括与目标变量高、低相关的特征.结果表明:构建的热轧带钢终轧温度随机森林预报模型在测试集上预测值的最大误差在15℃以内,具有较好的回归效果和泛化能力,满足热轧现场带钢终轧温度预报精度的要求.
Abstract
The complex boundary conditions and difficulty in parameter prediction in the precision rolling zone limit the accuracy of tradi-tional online hot rolling temperature prediction for strip steel.Therefore,in order to improve the accuracy of final rolling temperature predic-tion,a data-driven modeling approach using random forest was adopted,and taking forty-three characteristic factors that affect the final roll-ing temperature as the input values for the data-driven final rolling temperature prediction model.Then,the imbalanced datasets such as changing specifications were processed by a hybrid algorithm of NCL and SMOTE,and the random feature selection of the decision tree in-cluded features that were highly or lowly correlated with the target variable.The results show that the constructed random forest prediction model for the final rolling temperature of hot-rolled strip has a maximum prediction value error of within 15℃on the test set and good regres-sion effect and generalization ability,meeting the accuracy requirements for the final rolling temperature prediction of hot-rolled strip on site.
关键词
热轧/终轧温度/随机森林/数据驱动模型/非平衡数据集Key words
hot rolling/finish rolling temperature/random forest/data-driven model/non-equilibrium dataset引用本文复制引用
出版年
2025
NETL
NSTL
万方数据