基于LSTM的连铸异钢种混浇模型开发  

Development of a mixing model based on LSTM for heat transition of different steel grade in continuous casting process

作　　者：高宇 张瑞忠 曹金帅 张庆宇 刘宏春 年保国 李杰 孙剑 GAO Yu;ZHANG Ruizhong;CAO Jinshuai;ZHANG Qingyu;LIU Hongchun;NIAN Baoguo;LI Jie;SUN Jian(HBIS Material Technology Research Institute Process Technology Collaborative lnnovation Centre,Shijiazhuang 050000,Hebei,China)

机构地区：[1]河钢材料技术研究院工艺技术协同创新中心,河北石家庄050000

出　　处：《连铸》2024年第4期77-84,共8页Continuous Casting

摘　　要：为了进一步降低板坯异钢种连浇生产过程中的混浇坯长度,针对连铸生产过程中动态控制及对成分混合的影响,开展了精确预测混合过程及过渡坯划分问题的研究。通过工业试验分析了混浇生产过程中间包余钢量、拉速对成分混合的影响,发现在拉速、规格相同的条件下中间包余钢量越小,开浇后5 m铸坯位置下混合成分越接近后一炉成分,且其为主要影响因素;中间包余钢量、规格相同的条件下拉速越小,开浇后5 m铸坯位置下混合成分越接近后一炉成分,且其为次要影响因素。通过收集连铸机生产数据并结合数值模拟结果,建立了以LSTM神经元网络为基础的时间序列模型,通过模型离线计算,发现在控制混浇过程中中间包余钢量的平均值相等情况下,前期控制较低的中间包余钢量可加快中间包钢水的混合过程。模型在线布置应用结果表明,平均预测偏差为3.69%,精度满足实际生产需要。To further reduce the mixing length of grades transition in the slab casting process,research on accurate prediction of mixing process and transition slab division had been carried out in view of the time-varying control and its influence on the composition mixing in the continuous casting process.Through the industrial test,the influence of the amount of remaining steel in the tundish and the casting speed on the composition mixing during the transition casting process had been investigated.It has been found that with the same casting speed and section size the smaller the amount of remaining steel in the tundish is,the closer the mixed composition is to the composition of the subse-quent heat in the position of 5 m after new heat opened,which is the main influencing factor,and with the same re-sidual tundish weight and section size the smaller the casting speed is,the closer the mixed composition is to the composition of the subsequent heat in the position of 5 m after new heat opened,which is the secondary influencing factor.By collecting the production data of continuous casting process and combining the numerical simulation re-sults,a time series model based on LSTM neuron network has been established.Through the offline calculation of the model,it is found that the lower residual steel in the tundish in the early stage can accelerate the mixing process in the case of equal average tundish weight during mixing process.Deployed model online results show that the aver-age prediction deviation is 3.69%,and the accuracy meets the actual production needs.

关 键 词：异钢种连浇 动态控制 混浇模型 神经元网络 LSTM 

分 类 号：TF777[冶金工程—钢铁冶金]