基于自适应模糊聚类的LM-BPNN网络故障诊断  

LM-BPNN Based on Adaptive Fuzzy Clustering Algorithm to Network Fault Diagnosis

作　　者：蒋静芝[1] 孟相如[1] 张立[1] 

机构地区：[1]空军工程大学电讯工程学院,陕西西安710077

出　　处：《微电子学与计算机》2010年第7期153-156,共4页Microelectronics & Computer

基　　金：陕西省自然科学基金项目(SJ08F14)

摘　　要：提出一种综合运用模糊聚类和神经网络改进算法的网络故障诊断方法.根据网络故障特征数据量大、且存在冗余和冲突的特点,基于模糊聚类思想,提出了以聚类中心为核、自适应半径来优选样本的数据预处理方法.在进行故障特征的学习训练时,针对BP神经网络用于网络故障诊断时训练次数多、收敛慢和易振荡的局限性,使用结合了Levenberg-Marquardt的改进算法.理论分析和实验结果表明,文中提出的网络故障诊断方法能达到诊断更快速、更准确的效果.Owing to the fact that the characteristic data of networks are huge and have redundancy and conflict each other,a method of data pretreatment based on fuzzy clustering is proposed. The center of each kind is used to be a core,with adapting the radius to sample from training samples. During learning of the characteristic data,Levenberg-Marquardt algorithm is imported in order to resolve the weaknesses of BP,such as large training times,slow convergence and incidental surge. Theoretical analysis and experimental results show that this method is efficient for network fault diagnosis.

关 键 词：网络故障诊断 模糊聚类 自适应核区 LM-BPNN 

分 类 号：TP306[自动化与计算机技术—计算机系统结构] TP393[自动化与计算机技术—计算机科学与技术]