基于聚类思想的转炉终点碳含量预测方法  

Prediction method of end point carbon contentof converter based on clustering idea

作　　者：贺东风[1] 黄涵锐 He Dongfeng;Huang Hanrui(University of Science and Technology Beijing)

机构地区：[1]北京科技大学冶金与生态工程学院

出　　处：《冶金能源》2022年第6期29-34,共6页Energy For Metallurgical Industry

基　　金：国家重点研发计划资助项目(2017YFB0304001)。

摘　　要：针对转炉冶炼复杂、冶炼策略多,进而影响转炉终点预测模型命中率的问题,文章提出利用基于SNN_AP聚类算法与BPNN的转炉终点预测方法,最终将数据分为9类,总命中率在±0.03%内达到82.27%,较未分类时提升了约7个百分点。证明了聚类算法处理复杂炉况的有效性,为提升转炉预测模型精度提供一种重要思路。Aiming at the problems of complex converter smelting and many smelting strategies,which affect the hit rate of converter end-point prediction model,this paper proposes a converter end-point prediction method based on SNN_AP clustering algorithm and BPNN.Finally,the data are divided into 9 types of furnace conditions,and the total hit rate is 82.27%within±0.03%,which is about 7 percentage points higher than that without classification.The effectiveness of clustering algorithm in dealing with complex furnace conditions is proved,which provides an important idea for improving the accuracy of converter prediction model.

关 键 词：转炉终点预测 碳含量 共享最近邻 近邻传播 BP神经网络 

分 类 号：TF713[冶金工程—钢铁冶金]