摘要
为了提高RBF网络对时间序列的预测精度,引入软竞争ART,提出了基于软竞争机制的ARTRBF模型。在传统RBF网络的基础上,自适应控制生成隐含层节点的数目,并在第一阶段的学习中引入基于相似度的软竞争机制。与硬竞争ART-RBF网络相比,软竞争机制的采用,使得隐含层的每个节点都能参与对样本的学习,提高了节点的利用率,并且减少对预测精度有着重要影响的类间混叠处样本的误分。使用Matlab对Mackey-Glass时间序列进行仿真,并预测某轴承性能退化情况,结果表明该软竞争算法可以在一定程度上提高预测精度。
A prediction model which combines soft ART and RBF is proposed to improve the prediction accuracy of RBF.Based on traditional RBF network,the neurons in hidden layer is generated automatically by setting a vigilance parameter.A soft competition based on similarity is also introduced to this model.Compared with traditional ART-RBF network based on hard competition,the adoption of soft competition in learning process,every neuron will be used to adjust the center vectors; and it can also reduce the misclassification of similar samples that are very important to the prediction accuracy.This algorithm is proved to be efficient by the simulation of Mackey-Glass time series and prediction of bearing degradation.
出处
《武汉理工大学学报(信息与管理工程版)》
CAS
2016年第4期503-507,共5页
Journal of Wuhan University of Technology:Information & Management Engineering
基金
国家自然科学基金项目(51405353)
关键词
ART
软竞争
RBF
轴承
预测
ART
soft competition
RBF
bearing
prediction