摘要: 介绍一种新的多变量数据预处理方法——正交信号修正(OSC)法,提出一种OSC与NIPALS算法结合的O-PLS算法,将该方法用于确定BP网络的基本结构,即确定BP网络的隐层数、节点数及其初始权值,由此建立了O-PLS-BP网络模型。并将它与标准的BP网络模型、PLS-BP网络模型进行比较。仿真结果表明,该算法的拟合和预测效果较好。
关键词:
正交信号修正法,
O-PLS算法,
初始权值,
BP网络
Abstract: A preprocessing method for multivariate data called Orthogonal Signal Correction(OSC) is introduced. This method combines O-PLS algorithm with nonlinear iterative partial least-squares algorithm. This algorithm is used to a fundamental structure of BP network. This paper proposes a novel BP network model based on O-PLS algorithm. The weights initialization and the member of hidden nodes are set by applying the loading weights and factor numbers of O-PLS algorithm. Performances of the BP, PLS, and PLS-BP are analyzed and compared for instance. The result shows that the PLS-BP has better fitting and forecast than BP, PLS.
Key words:
Orthogonal Signal Correction(OSC),
O-PLS algorithm,
weights initialization,
Back Propagation(BP) network
中图分类号:
张艳粉;刘琼荪. 基于正交信号修正法的BP网络模型及其应用[J]. 计算机工程, 2008, 34(5): 228-230.
ZHANG Yan-fen; LIU Qiong-sun. Back Propagation Network Model and Its Application Based on Orthogonal Signal Correction[J]. Computer Engineering, 2008, 34(5): 228-230.