多元线性回归的预测建模方法[J].pdfVIP

2007 4 April 2007 33 4 Journal ofBeijingUniversity ofAeronautics andAstronautics Vo.l 33 No1 4 王惠文 孟 洁 ( , 100083) : 根据历史的样本数据, 建立多元线性回归的预测模型; 从而在不需要 来样 本数据的情况下, 预测 来时刻多元线性回归模型中的回归参数, 以及主要的模型精度评估指 标. 对多元线性回归模型参数的预测, 转化为对其变量集合的增广矩阵的叉积阵的预测. 对叉 积阵进行谱分解, 利用高维群点主轴旋转的预测建模方法, 通过 Givens变换得到特征向量矩 阵的转角值, 对自由取值的转角以及特征值建立预测模型. 仿真实验例示了该方法的主要计算 步骤; 计算结果显示, 利用本模型得到的拟合值精度较高, 预测值真实可信. 最终计算结果和实 验结果吻合较好, 表明这种方法可以用于分析和预测众多领域中因变量对自变量的回归关系 问题. : 回归分析; 建模; 预测; 评估 : O 212. 4 : A : 1001259 5(2007) 0420500205 P re d ic tive m ode ling on m u ltiva r ia te lin ea r reg re s s ion W angHu wi en Meng Jie ( School of Econom ics andManagement, BeijingU n iversity ofAeronautics and Astronautics, Beijing 100083, Ch ina) Ab s tra c t: Based on the h istorical data, predictive method ofmu ltivariate linear regression modelwas dis2 cussed, where both futuremu ltivariate regression parameters and the performance evaluation statisticswere es2 tmi ated without future data. Prediction to the regression parameterswas converted to predict cross productm a2 trix of the variab le augmented matrix. By app lying spectral decomposition, cross product matrix was decom2 posed of eigenvectors and eigenvalues. Predictivem ethod of orthonomalmatrix based on rotations of principal axes was adopted to predict eigenvector matr ix, where rotated angles from G ivens transform ation were ob2 tained, and regular predictive modelswere bu ilt on both the angles and eigenvalues, respectively. The experi2 mental smi ulation illustratesmain computational procedures of the predictivemodel. Besides, the resu lts show a high precise of the fitted values and a statistical validity of the predictive