计算机与应用化学
計算機與應用化學
계산궤여응용화학
COMPUTERS AND APPLIED CHEMISTRY
2013年
7期
793-796
,共4页
独立成分回归%多模型共识%紫外光谱%定量分析
獨立成分迴歸%多模型共識%紫外光譜%定量分析
독립성분회귀%다모형공식%자외광보%정량분석
independent component regression%consensus modeling%ultraviolet spectroscopy%quantitative analysis
将独立成分分析用于紫外光谱定量分析,结合多模型共识的基本思想,建立了共识独立成分回归方法。从训练集随机取样建立一系列独立成分回归模型,选取其中性能较好的部分模型作为成员模型,并用这些成员模型预测未知样品。用该方法对苯甲酸、苯胺及苯酚3组分水溶液的紫外光谱进行分析,并与单模型偏最小二乘法了进行比较。结果PLSR对独立测试集中3种组分进行50次重复预测的平均RMSEP分别为2.349,7.413和1.605,RMSEP的标准偏差分别为1.781,2.918和1.266;而本方法重复50次预测的平均RMSEP分别为1.633,3.390和1.496,RMSEP的标准偏差分别为6.642×10-3,6.573×10-2和4.484×10-2。可见,共识独立成分回归所建立的模型更加稳健和可靠,预测的准确性也明显提高。
將獨立成分分析用于紫外光譜定量分析,結閤多模型共識的基本思想,建立瞭共識獨立成分迴歸方法。從訓練集隨機取樣建立一繫列獨立成分迴歸模型,選取其中性能較好的部分模型作為成員模型,併用這些成員模型預測未知樣品。用該方法對苯甲痠、苯胺及苯酚3組分水溶液的紫外光譜進行分析,併與單模型偏最小二乘法瞭進行比較。結果PLSR對獨立測試集中3種組分進行50次重複預測的平均RMSEP分彆為2.349,7.413和1.605,RMSEP的標準偏差分彆為1.781,2.918和1.266;而本方法重複50次預測的平均RMSEP分彆為1.633,3.390和1.496,RMSEP的標準偏差分彆為6.642×10-3,6.573×10-2和4.484×10-2。可見,共識獨立成分迴歸所建立的模型更加穩健和可靠,預測的準確性也明顯提高。
장독립성분분석용우자외광보정량분석,결합다모형공식적기본사상,건립료공식독립성분회귀방법。종훈련집수궤취양건립일계렬독립성분회귀모형,선취기중성능교호적부분모형작위성원모형,병용저사성원모형예측미지양품。용해방법대분갑산、분알급분분3조분수용액적자외광보진행분석,병여단모형편최소이승법료진행비교。결과PLSR대독립측시집중3충조분진행50차중복예측적평균RMSEP분별위2.349,7.413화1.605,RMSEP적표준편차분별위1.781,2.918화1.266;이본방법중복50차예측적평균RMSEP분별위1.633,3.390화1.496,RMSEP적표준편차분별위6.642×10-3,6.573×10-2화4.484×10-2。가견,공식독립성분회귀소건립적모형경가은건화가고,예측적준학성야명현제고。
The independent component analysis was used for ultraviolet spectrum quantitative analysis, and a consensus independent component regression (cICR) method was proposed based on the basic idea of consensus modeling. A series of ICR models was built on training subsets which were constructed by random sampling from the training set;the models with high performance were selected as member models, and were used for prediction. The cICR was used for modeling on ultraviolet spectroscopic data which derived from a series of tri-component aqueous solution of benzoic acid, aniline and phenol. Meanwhile, the method was compared with the single-model partial least squares regression (PLSR). As results, the single-model PLSR obtained 2.349, 7.413 and 1.605 of mean RMSEP on 50 repeat prediction for the three components on the independent test set, the standard deviation of the RMSEPs were 1.781, 2.918 and 1.266, respectively. While cICR obtained 1.633, 3.390 and 1.496 of mean RMSEP and 6.642×10-3, 6.573×10-2 and 4.484×10-2 of correspond standard deviations. The results shown that the models built by cICR are more steady and reliable;and the prediction results are more accurate than single-model PLSR.