摘要
为了更好地确定偏最小二乘法模型的主成分数,提出一种传统偏最小二乘法和多主成分数偏最小二乘法相结合构建复合偏最小二乘模型的方法。给出了预测时两种样品相似度的计算方式:直接距离法和性质得分距离法。分别采用复合偏最小二乘法和传统偏最小二乘法对煤炭的全硫、灰分、热值和碳含量进行建模预测,比较传统偏最小二乘法和多主成分数偏最小二乘法建模过程中的相关系数和交互验证均方根误差,采用复合偏最小二乘模型对验证集样品预测时,计算了不同相似度计算方式下不同样品间距离算法的预测均方根误差,并同传统偏最小二乘法预测均方根的误差进行比较,结果表明:复合偏最小二乘法建模比传统偏最小二乘法建模有更强的适应性,能够提高预测的准确性。
In order to determine the number of PLS(partial least squares)principal component,the compound PLS method developed from multi-principal component number PLS and traditional PLS was proposed and both direct distance method and property score method for sample similarity calculation were given;The total sulfur content,ash content,calorific value,carbon content of coal were predicted with compound PLS and traditional PLS,and the correlation coefficient(R2)and RMSECV(root mean square error of cross validation)in MPLS and PLS were compared;RMSEP(root mean square error of prediction)in different similarity calculation methods and algorithms of the distance between the samples were calculated and compared with that of traditional PLS while having the samples predicted with compound PLS model.The results indicated that the compound PLS model outperforms the traditional one in adaptability and accuracy.
出处
《化工自动化及仪表》
CAS
2012年第2期178-181,189,共5页
Control and Instruments in Chemical Industry