基于梯度提升算法的轧后冷却区内带钢运行速度预测  

Prediction of strip running speed in cooling zone after rolling based on gradient boosting algorithm

作　　者：丁鹿西 龚殿尧[1] 田宝钱 郝秋宇 徐建忠[1] 王秋娜[2] DING Luxi;GONG Dianyao;TIAN Baoqian;HAO Qiuyu;XU Jianzhong;WANG Qiuna(State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China;Hot Rolling Plant,Shougang Qian'an Iron&Steel Company,Qian'an 064404,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]首钢股份公司迁安钢铁公司热轧厂,河北迁安064404

出　　处：《冶金自动化》2023年第4期62-69,99,共9页Metallurgical Industry Automation

摘　　要：卷取温度是影响成品带钢组织性能的重要指标之一,在影响卷取温度控制精度的诸多因素中,带钢运行速度尤为重要。针对生产过程中轧后冷却区内带钢运行速度因受到终轧温度控制模型的影响变得难以预测的问题,结合大量现场生产数据,采用基于数据驱动的梯度提升决策树算法(gradient boosting decision tree,GBDT)预测带钢样本点在精轧出口的速度,并与随机森林模型(random forest,RF)和支持向量机回归模型(support vector regression,SVR)进行对比分析。模型对比结果表明,GBDT模型的平均绝对误差EMA(mean absolute error,MAE)最小,为0.07179,预测精度和泛化性能均好于对比模型,实现了轧后冷却区内带钢运行速度的高精度预测,为前馈控制提供了有效的参考数据,对卷取温度的高精度控制具有重要意义。The coiling temperature is one of the important indicators affecting the microstructure and properties of the finished strip.Strip running speed is the important factor in high precision control of coiling temperature.In response to the problem that the running speed of strip steel in cooling zone after rolling during production becomes difficult to predict due to the influence of the final rolling temperature control model,a large amount of on-site data was combined and the gradient boosting decision tree(GBDT)algorithm based on data-driven approach was used to predict the speed of strip steel samples at the exit of the finishing mill.The model was compared with random forest(RF)and support vector regression(SVR)models.The comparison results show that the GBDT model has the smallest average absolute error EMA(mean absolute error,MAE)of 0.07179,and the prediction accuracy and generalization performance are better than the comparison model.The high-precision prediction of the strip running speed in the cooling zone after rolling is realized,which provides effective reference data for the feed-forward control and is of great significance to the high-precision control of the coiling temperature.

关 键 词：热轧带钢 轧后冷却 速度预测 时间-速度-位置曲线 梯度提升 

分 类 号：TG335.56[金属学及工艺—金属压力加工]