摘要
为了选择与预报量有密切关系的因子和有代表性的样本，分别采用投影寻踪回归（ＰＰＲ）和ＢＰ神经网络对因子和样本进行优选。实例分析表明：优选出的样本保留了全部样本的大部分信息；用优选出的因子建模可达到与用全部因子建模相近的拟合和预报效果。
In order to select factors closely related to the prediction and representative samples, the factors and samples are optimized using PPR and B P neural network respectively.The analysis of examples shows that after optimization the samples maintain the most information and the modelling of factors has the close effects.
