一种基于主元选择的偏最小二乘回归方法.pdf

31 16 2005 8 Vol.31 16 Computer Engineering August 2005 · · 2005 A TP39 710038 A Partial Least-squares Regression Method Based on Principal Variables Selection LI Shouan, ZHANG Hengxi, GUO Jilian, MEN Ke (Dept. of Aircraft and Engine Engineering, Engineering Institute, AFEU, Xi’an 710038) AbstractTo analyze multivariate data with few observations more efficiently, a new partial least-squares regression method based on principal variables selection is proposed, and the basic theories and calculation process of the method is explained. The method selects principal variables in data samples by correlative coefficients matrix, and then regresses the principal variables by partial least-squares regression. Compared with partial least-squares regression method, the method has higher precision and fewer regression steps to eliminate the multicollinearity in data. It is illustrated that partial least-squares regression method based on principal variables selection is more effective and accurate in analyzing multivariate data with few observations than partial least-squares regression method. Key wordsPrincipal variables selection; Partial least-squares regression; Multivariate data with few observations; Multicollinearity y y x i F p -1 F F [3] F j S.Wold C.Albano p-2 F 1983