摘要
神经网络已广泛应用于设备的故障诊断中。鉴于神经网络的性能和诊断能力与网络的拓扑结构和学习算法有着密切的联系,本文研究并实现了在故障诊断应用中能进行自构形的神经网络模型,改善了全局收敛性和节点总体饱和度,并较好地应用于变压器故障诊断实践中。
Artificial Neural Network(ANN)has been extensively used to diagnose equipment faults. ANN抯 performance and diagnostic capability have relation with their topological structure and training algorithm. This paper studies and practises ANN抯 auto-structural model in diagnosing faults. By auto-structuring in training and importing punish-function arithmetic to resist overfitting and improve the overall convergence and node collectivity saturation, this model has been applied in transformer faults detection and indicates good effect.
出处
《电工技术学报》
EI
CSCD
北大核心
2004年第9期77-81,共5页
Transactions of China Electrotechnical Society
关键词
神经网络
自动构形
变压器
故障诊断
Artificial neural network,auto-structure,transformer,fault diagnosis
