东北大学学报(自然科学版) ›› 2024, Vol. 45 ›› Issue (10): 1369-1378.DOI: 10.12068/j.issn.1005-3026.2024.10.001

• 信息与控制 •    

基于Stacking集成的RF-ET-KDE烧结过程物理指标区间预测模型

康增鑫, 陈进朝, 王金杨, 吴朝霞()   

  1. 东北大学秦皇岛分校 控制工程学院,河北 秦皇岛 066004
  • 收稿日期:2023-05-24 出版日期:2024-10-31 发布日期:2024-12-31
  • 通讯作者: 吴朝霞
  • 作者简介:康增鑫(1999-),男,河北石家庄人,东北大学硕士研究生
    吴朝霞(1969-),女,浙江嘉兴人,东北大学教授.
  • 基金资助:
    河北省教育厅科学技术研究项目(BJ2021099)

Interval Prediction Model of RF-ET-KDE Sintering Process Physical Index Based on Stacking Integration

Zeng-xin KANG, Jin-chao CHEN, Jin-yang WANG, Zhao-xia WU()   

  1. School of Control Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China.
  • Received:2023-05-24 Online:2024-10-31 Published:2024-12-31
  • Contact: Zhao-xia WU
  • About author:WU Zhao-xia,E-mail: ysuwzx@126.com

摘要:

由于烧结过程中存在众多不确定性因素,使得机理分析和点预测结果的可靠性不足.基于此提出随机森林-极限树-核密度估计(random forest-extreme tree-kernel density estimation,RF-ET-KDE)算法对物理指标(粒度、水分)进行区间预测.首先,采用数据预处理和特征选择操作筛选出最适合建模的特征变量.其次,使用基于Stacking的RF-ET算法对指标进行点预测,该算法使得模型有较高的准确性和泛化性.然后,采用KDE算法计算指标的预测误差,得到了一定置信水平下的分布区间和区间预测结果.最后,用所建模型与其余组合模型进行对比.结果表明,RF-ET算法有较高的点预测效果,KDE算法可以很好地量化指标的误差,可以得到较高可靠度的区间预测结果.

关键词: 烧结过程, 随机森林-极限树, 核密度估计, 物理指标, 区间预测

Abstract:

Due to the many uncertainties in the sintering process, the reliability of mechanism analysis and point prediction results is insufficient. Therefore, a random forest-extreme tree-kernel density estimation (RF-ET-KDE) algorithm is proposed to realize interval predictions for physical indicators, such as particle size and moisture. Firstly, data preprocessing and feature selection operations are adopted to screen out the most suitable feature variables for modeling. Secondly, the RF-ET algorithm based on Stacking is utilized to realize point predictions for the indicators. This algorithm makes the model with higher accuracy and generalization, and then the KDE algorithm is adopted to calculate the prediction error of the indicator. The distribution interval and interval prediction results under a certain confidence level are obtained. Finally, the proposed model is compared with the other combined models. The results show that the RF-ET algorithm has higher point prediction accuracy, and the KDE algorithm can quantify the error of the indicator very well, so that a higher credibility interval prediction result can be obtained.

Key words: sintering process, random forest-extreme tree (RF-ET), kernel density estimation (KDE), physical index, interval prediction

中图分类号: