基于贝叶斯优化XGBoost模型的热轧中厚板终冷温度预测  

Prediction of final cooling temperature for hot rolled medium and heavy plate based on Bayesian optimization XGBoost model

作　　者：杜岩 王义铭 张田 田勇[1] 王丙兴[1] DU Yan;WANG Yiming;ZHANG Tian;TIAN Yong;WANG Bingxing(State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China;Department of Mechanical Engineering,Shenyang Jianzhu University,Shenyang 110168,China)

机构地区：[1]东北大学轧制技术及连轧自动化国家重点实验室,辽宁沈阳110819 [2]沈阳建筑大学机械工程学院,辽宁沈阳110168

出　　处：《轧钢》2022年第6期91-98,共8页Steel Rolling

基　　金：国家重点研发计划项目(2018YFB1701600)。

摘　　要：轧后冷却是影响钢板组织性能的重要工序之一,其中终冷温度是冷却过程的关键控制参数。由于传统的温控数学模型与自学习模型精度较差,因此为了提高终冷温度预测的精度,采用贝叶斯优化XGBoost模型对终冷温度进行回归预测,以冷却速率、总流量、开启集管数和化学成分等15个变量作为模型的输入,终冷温度作为模型输出,通过BO-XGBoost模型预报终冷温度。同时,对比了BP神经网络模型、默认超参数XGBoost模型和BO-GBDT模型的回归效果。结果表明:BO-XGBoost模型的训练集和测试集具有较高的决定系数和较低的误差,说明模型具有很好的泛化性、鲁棒性和非线性拟合能力,相较于经典数学模型提高了终冷温度预测精度。Cooling after rolling is one of the important processes that affect the microstructure and properties of the plate, and final cooling temperature is the key control parameter of the cooling process. Since the accuracy of traditional temperature control mathematical models and self-learning models is poor, a Bayesian optimlzation XGBoost model was used for regression prediction of final cooling temperature in order to improve the accuracy of final cooling temperature prediction. 15 variables such as cooling rate, total flow rate, number of open headers, and chemical composition were used as the inputs to the model, and the final cooling temperature was used as the model output. The final cooling temperature was predicted by the BO-XGBoost model. Meanwhile, the regression effects of BP neural network model, XGBoost model of default hyperparameters and BO-GBDT model were compared. The result show that the training set and test set of BO-XGBoost model have higher decision cofficients and lower errors, which shows that the model has good generalization, robustness and nonlinear fitting ability and improves the final cooling temperature prediction accuracy compared with classical mathematical models.

关 键 词：中厚板 终冷温度 轧后冷却 贝叶斯优化 XGBoost模型 

分 类 号：TG335.11[金属学及工艺—金属压力加工]