引用本文:张海刚,张森,尹怡欣,张晓娟.高炉料面特征提取与聚类分析[J].控制理论与应用,2017,34(7):938~946.[点击复制]
Zhang Hai-gang,Zhang Sen,Yi Yi-xin,Zhang Xiao-juan.Feature extraction and clustering of blast furnace burden surface[J].Control Theory and Technology,2017,34(7):938~946.[点击复制]
高炉料面特征提取与聚类分析
Feature extraction and clustering of blast furnace burden surface
摘要点击 2790  全文点击 2027  投稿时间：2016-11-27  修订日期：2017-04-08
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2017.60903
  2017,34(7):938-946
中文关键词  高炉  料面  特征提取  谱聚类  特征匹配
英文关键词  blast furnace  burden surface  feature extraction  spectral clustering  feature matching
基金项目  国家自然科学基金项目（No.61333002，61673056，61671054）
作者单位E-mail
张海刚 北京科技大学自动化学院 zhang_gang1989@126.com 
张森* 北京科技大学自动化学院 zhangsen@ustb.edu.cn 
尹怡欣 北京科技大学自动化学院  
张晓娟 北京科技大学自动化学院  
中文摘要
      布料是高炉上部调节的主要方式，料面形状特征是指导高炉工长做出下一次布料决策的重要依据。本文通过分析高炉雷达实测料面数据，结合专家经验，提出了一种料面特征定义和提取的方法，提取出了能够表征料面形状的六个特征；然后将谱聚类算法引入到料面特征数据的聚类中，建立了料面特征模型库；最后将新的料面特征与模型库数据匹配，为后续高炉布料控制奠定研究基础。在仿真结果中采用真实高炉生产数据，验证了料面特征提取方法的有效性；仿真结果表明谱聚类算法相比传统的K-mean和模糊C均值聚类算法，具有收敛速度快，聚类性能高等特点，能够准确有效的建立料面特征模型库。
英文摘要
      Burden distribution is the main way to adjust the upper part of the blast furnace. Shape features of the burden surface are the important basis for guiding the blast furnace foreman to make the next burden distribution decision. In this paper, a new burden surface definition and extraction approach is proposed by analyzing the measured surface radars data and combining with expert experience, six features are extracted to characterize the shape of burden surface. The spectral clustering algorithm is then utilized to the extracted feature clustering problem in order to set up the standard feature model database of the burden surface. Finally, the new burden surface features data will be matched with the samples in the history model library. This work will lay the foundation for further study of the blast furnace burden distribution control. The simulation results show that this feature extraction method and the clustering algorithm is effective. Compared with the ordinary K-Mean and Fuzzy C-Mean clustering algorithms, the spectral clustering algorithm presents better performance with relatively faster convergent speed, which is more efficient and accurate to establish the history model library for the burden line.