基于ART2神经网络算法改进的研究  

Research of an Improved Algorithm Based on ART2 Neural Network

作　　者：吕秀江[1] 王鹏翔[1] 王德元[1] 

机构地区：[1]长春工业大学电气与电子工程学院,吉林长春130012

出　　处：《计算机技术与发展》2009年第5期137-139,共3页Computer Technology and Development

基　　金：吉林省科技计划项目(20030532)

摘　　要：ART2神经网络是按照自适应谐振理论建立的一种自组织、无监督的人工神经网络。通过分析经典自适应谐振神经网络聚类过程,针对传统ART2神经网络模型对分类的不确定性和网络权值模式漂移等不足,提出了基于算法改进的ART2神经网络模型。最后对改进的ART2神经网络进行了仿真,并与经典神经网络所做仿真的结果比较,验证了改进的ART2神经网络结构大大提高了分类的正确率,有效改善了模式漂移现象,降低了空间存储消耗。ART2 neural network is a self- organized and unsupervised artificial neural network constructed from adaptive resonance theory. An improved ART2 neural network clustering algorithm was proposed by analyzing such the shortcomings of classical ART2 neural network clustering algorithm, such as the indefinability of classification and network pattern drifting etc. Finally provide the relevant algorithm as well as the comparison with classical ART2, the simulation and experimental results verified the improved ART2 neural network has enhanced the validity of classification greatly,reformed pattern drifting and greatly economized the networks＇ memory capacity effectively.

关 键 词：ART2神经网络 模式漂移 分类 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]