摘要
针对小波网络隐层节点的选择问题,结合QR分解提出一种新的正交化选择方法。在节点选择过程中,预先对候选节点集合进行分块操作,在每个子块里采用QR分解与误差减小率(error reduction ratio,ERR)参数相结合的方法进行选择。该方法进一步简化了小波网络的计算量,并且保持网络较高的计算精度。将该方法应用于采用变量解析(the analysis of variance,ANOVA)扩展的小波网络,分别对Mackey-Glass方程和年太阳黑子数据进行一步预测,结果表明,提出的正交化选择方法在简化计算的同时,能较好地提取系统特征,并显示出较好的预测结果。
A new algorithm for the node selection was proposed for the hide layer of wavelet networks, This new algorithm divided the original node group into several parts to avoid comparing among the poor nodes and used QR decomposition to select the optimum nodes. The burden of the heavy calculation with a good precision was relieved. Applying this algorithm to the wavelet network with the analysis of variance （ANOVA） expansion made one-step-ahead predictions, respectively, for the Mackey-Glass delay-differential equation and the annually sunspot data set, The results show that the new algorithm can catch the characteristics of the systems and has a good performance,
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2007年第17期3899-3902,3965,共5页
Journal of System Simulation
基金
国家自然科学基金项目(60674073
60374064)
关键词
小波网络
QR分解
非线性系统
一步预测
wavelet network
QR decomposition
non-linear system
one-step-ahead prediction
