基于神经网络算法的热轧钢板凸度预报  

Profile-Predicting for Hot-Rolled Steel Sheets Based on Neural Network Algorithms

作　　者：张守峰[1] 王笑辰 赵健[3] 赫竟彤 宋君 马晓国 ZHANG Shoufeng;WANG Xiaochen;ZHAO Jian;HE Jingtong;SONG Jun;MA Xiaoguo(Hot Rolled Strip Steel Mill of Angang Steel Co.,Ltd.,Anshan 114021,Liaoning,China;Ansteel Beijing Research Institute Co.,Ltd.,Beijing 102209,China;Silicon Steel Busniness Division of Angang Steel Co.,Ltd.,Anshan 114009,Liaoning,China;School of Materials Scienceand Engineering,Northeastern University,Shenyang 110819,Liaoning,China)

机构地区：[1]鞍钢股份有限公司热轧带钢厂,辽宁鞍山114021 [2]鞍钢集团北京研究院有限公司,北京102209 [3]鞍钢股份有限公司硅钢事业部,辽宁鞍山114009 [4]东北大学材料科学与工程学院,辽宁沈阳110819

出　　处：《鞍钢技术》2024年第5期18-25,共8页Angang Technology

摘　　要：提出了一种基于Elman神经网络的预测模型,并结合布谷鸟算法(CS)对网络的初始权值和阈值进行优化,以提高钢板凸度的预测精度。通过收集鞍钢股份有限公司热轧带钢厂的生产数据并进行预处理后训练模型,结果表明,本文提出的CS-Elman模型的平均绝对值误差(MAE)为1.3693、均方误差(MSE)为3.0843、平均绝对百分比误差(MAPE)为3.9025%,决定系数R2为0.95123,以上指标均较原始Elman算法表现出明显的提升。该预测模型能够有效挖掘生产数据中的潜在规律,为钢板凸度的精准预测提供了一种有效的解决方案,对优化热轧过程和提升产品质量具有重要的实际应用价值。The prediction model based on the Elman neural network was proposed and then the initial weight values and threshold values in terms of networks were optimized by syncretizing the Cuckoo Search(CS)algorithm so as to improve the prediction accuracy of the profiles of steel sheets.After collecting production data from Hot Rolled Strip Steel Mill of Angang Steel Co.,Ltd.and pre-processing these data,the model was exercised experimentally,the experimental results showed that the proposed CS-Elman model was characterized by having the mean absolute error(MAE)of 1.3693,mean square error(MSE)of 3.0843,mean absolute percentage error(MAPE)of 3.9025%,and coefficient of determination R2 of 0.95123.All these indicators showed signifi-cant improvement compared to the original Elman algorithm.This prediction model can effectively extract underlying laws from production data,which provided an effective solution for accurately predicting the profiles of steel sheets.So this model had significant practical application values for optimizing hot rolling processes and improving product quality.

关 键 词：板凸度预测 神经网络 布谷鸟算法 数据驱动 

分 类 号：TG335[金属学及工艺—金属压力加工]