基于Bagging的多模型钢水温度预报  

Multi-model prediction of molten steel temperature based on Bagging

作　　者：田慧欣[1] 毛志忠[1,2] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110004 [2]东北大学流程工业综合自动化教育部重点实验室,沈阳110004

出　　处：《控制与决策》2009年第5期687-691,共5页Control and Decision

基　　金：国家自然科学基金项目(60374003)

摘　　要：针对LF(Ladle Furnace)冶炼特点及现有钢水温度预报方法存在的不足,提出一种基于Bagging的多模型预报方法.该方法利用Bagging思想,有效地将智能方法与机理方法相结合,在提高智能方法预报精度的同时使得机理模型与智能模型的优势得到互补,克服了单纯机理模型参数获得不够准确以及单纯智能模型缺乏工艺指导的不足.另外,该方法还对Bootstrap采样后的子训练数据集进行了主成分分析(PCA),有效地克服了集成算法固有的子学习机精度与数据集多样性不能兼顾的问题.A multi-model prediction method based on Bagging is proposed for ladle furnace, aiming at the characters of ladle furnace metallurgic process and the disadvantages of traditional prediction methods. In this method, the Bagging method is used to combine the intelligent method with the mechanism method efficiently. The new method can improve the accuracy of intelligent model and overcome the disadvantages of intelligent model and mechanism method. Furthmore the principal component analysis （PCA） is ingeniously integrated into the new multi-model method for the Bootstrapped data. The problem of ensemble algorithm about ensuring the accuracy of learner and the diversity of data set simultaneously is solved efficiently by using PCA.

关 键 词：BAGGING 多模型 主成分分析 机理模型 智能模型 极限学习机 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置]