基于神经网络的非线性PCA方法

由于普通的主元分析（PCA）方法无法提取数据中的非线性相关特性，本文提出了一种基于神经网络的非线性PCA（NIPCA）方法，不仅提取了高维原始数据的线性信息还能提取非线性信息。在此基础上进一步提出了样本中显著误差及劣点的检测方法，从而支持对其进行合理剔除或是修正，仿真试验表明它能有效地减小误差点对网络训练精度的影响，大大增强了算法的鲁棒性。

A Nonlinear PCA Method Based on Neural Networks

Because the ordinary linear PCA can't extract nonlinear features in data,an approach of nonlinear principal component analysis(NLPCA)based on the auto-associative neural network is presented in this paper,which can extract not only the linear features but also the nonlinear ones in high dimensional data.Further more,a method of detecting the outliers and the gross errors was presented based on this NLPCA algorithm,and those outliers and errors can be eliminated or revised rationally.The simulation results show that this method successfully reduces the errors,effectively improves the precision of the prediction and the robustness of the NLPCA algorithm.