摘要
在自组织神经网络基本学习训练算法基础上,本文从初始连接权值的设置、距离函数的定义、学习速率的自适应调节、收敛准则的确定以及分类模板的调整等五个方面分析如何改进传统的自组织神经网络算法,设计了一种改进的自组织神经网络结构,以电力变压器的故障诊断为例,实验结果表明,该方法能够取得较好的故障诊断结果,具有一定的实际应用价值。
The basic learning algorithm of self - organizing feature map neutral network is introduced in the paper, and we talk about how to improve the basic algorithm from five perspectives： initialization of the weight vectors, the definition of distance between the weight vectors, the self - adaptation of learning rate, the criterion of convergence and the fine - tuning of classification. Then, an improved algorithm for self - organizing feature map neutral network was presented, and applied for fault diagnosis of a power transformer. The algorithm has such advantages as high speed , high correct and better results, and the effectiveness and meaning of the algorithm has been verified through the application of the example.
出处
《微计算机应用》
2007年第11期1127-1131,共5页
Microcomputer Applications
基金
国家重大基础研究计划(973)前期研究专项项目(2002CCC01900)
关键词
自组织神经网络
改进方法
故障诊断
Self - Organizing Neural Network, Improved Algorithm, Fault Diagnosis
