摘要
依据多分辨分析和正交小波分解理论,建立由正交小波函数和正交尺度函数共同作为神经网络的激励函数的正交小波网络,并充分利用二者互补的特性,给出正交小波网络的分层、递阶学习算法,这使得正交小波网络在逼近函数的过程中不存在局部最小的问题,而且可以建立网络结构与逼近精度之间的明确关系;建立了我国人口预测的正交小波网络模型并验证了模型的实际有效性.
Based on multiresolution analysis and orthonormal wavelet decomposition theory, orthonormal wavelet network is built, with both orthonormal wavelet and orthonormal scaling functions being activation functions. By making full use of complementarity of two activation functions, layered and hierarchical learning algorithms for the network are given, which makes orthononnal wavelet network approach to desired solution without being trapped in local minima and is able to build definite relationship between network＇s stmcture and approximation accuracy. Furthermore, case study .is carried out to demonstrate application of orthonormal wavelets networks to population forecasting of our country and result is encouraging.
出处
《系统工程学报》
CSCD
北大核心
2006年第2期196-200,共5页
Journal of Systems Engineering
基金
国务院侨办科研基金资助项目(04QSK05)
华侨大学博士基金资助项目(05BS104)
关键词
正交小波
尺度函数
多分辨分析
经济预测
orthonormal wavelet
scaling functions
multiresolution analysis
economics forecasting
