摘要
作为一种经典的回归建模方法,偏最小二乘法(partial least squares,PLS)已被广泛的应用于软测量建模中。但是,当建模数据混有较大噪声时,采用PLS模型的预测误差以及预测误差的方差都比较大。针对PLS方法的上述缺陷与不足,本文将迭代Bagging算法引入PLS回归建模中,形成迭代Bagging PLS算法(iterated Bagging PLS,IBPLS),该方法可以减少预测误差和预测误差的方差。仿真结果表明,与传统PLS方法相比,IBPLS减小预测误差约6%。
The partial least squares (PLS) regression method has been widely applied to soft sensing systems. However,the PLS predictions have large variances and biases when the original data includes noise.This paper describes a PLS regression method combined with an iterated bagging method which reduces the predictor variance and bias.Simulations show that this approach reduces the model output variance by about 6%.
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2008年第S2期1780-1784,共5页
Journal of Tsinghua University(Science and Technology)
基金
国家"八六三"高技术项目(2006AA04Z168)