摘要
ART-2A神经网络可以很好地应用于模式识别中的分类问题,但由于算法结构的不足,存在分类漂移现象,导致网络不稳定,严重影响了网络的工程应用。对标准ART-2A网络结构和算法过程进行了深入研究,分析了引起分类漂移的原因,提出了一种ART-2A神经网络改进算法,并通过故障自诊断实验进行网络稳定性和实用性验证。实验证明改进ART-2A神经网络能长期稳定工作,准确进行故障判别,实现系统的故障诊断自动化。
ART - 2A Neural Network is effective in dealing with classification problem of pattem recognition, but it suffers from classification drift because of the shortage in algorithm structure, which may cause network instability, and in turn severely affect the engineering application of the network. The standard ART- 2A network structure and algorithm are studied. And the causes of classification drift are analyzed. An improved ART- 2A algorithm is presented and is used in Fault Detection and Diagnosis（FDD） experiments for verification of network stability and usability. The Experiments showed that the improved ART - 2A algorithm can work steadily in a long term, and can distinguish fault correctly. The improved ART- 2A network is helpful for realizing automatic FDD.
出处
《电光与控制》
北大核心
2008年第10期84-88,共5页
Electronics Optics & Control
关键词
故障诊断
神经网络
ART-2A
Fault Detection and Diagnosis（FDD）
neural network
ART- 2A
作者简介
刘智荣（1983-），男，湖南永州人，硕士生，研究方向为测试技术及设备。
