摘要
为了提高复杂系统故障识别的精度和降低误报率,利用粗糙集理论、遗传算法、神经网络等计算智能方法的优势,提出一种基于计算智能技术融合的故障识别方法.针对原始样本数据的不确定性和不完备性,利用粗糙集对原始样本数据进行数据归一化、离散化、属性约简等预处理,求得能够覆盖原始数据特征的具有最大完备度的最小规则集.利用具有全局搜索能力的遗传算法直接训练反向传播神经网络的权值,将规则集作为网络输入,形成优化网络模型.采用该模型对预处理的各种状态故障特征向量进行分类决策,实现故障识别.通过电机轴承故障识别实验表明,该方法能够优化网络结构,提高故障识别速度和准确率.
A computational intelligence technology fusion method of fault detection was proposed in order to improve the detection precision and decrease the misinformation detection of complex system.The various approaches such as rough set,genetic algorithm and neural network were integrated to synthesize their merits for fault detection.According to the uncertainty and imperfection of the original sample data,the rough set was used to pretreat for the normalization of data,the discretization of continuous data and the attribute reduction in order to obtain the minimum fault feature subset.The genetic algorithm with the ability of strong global search was used to train the weights of back propagation neural network.The minimum reduced subset was inputted into the trained network to construct the fault detection model that can classify the pretreated fault feature vectors under certain states to realize the fault detection.The motor bearing experiment results show that the method can optimize the structure of neural network and improve the rate and precision of fault detection.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2010年第7期1298-1302,共5页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(60870009
60775028)
智能制造湖南省高校重点实验室开放课题资助项目(2009IM03)
关键词
计算智能
融合
故障识别
粗糙集
反向传播神经网络
遗传算法
属性约简
computational intelligence
fusion
fault detection
rough set
back propagation neural network
genetic algorithm
attribute reduction
作者简介
邓武（1976-），男，四川安岳人，博士生，从事计算智能研究．E—mail：dw7689@163．com
通信联系人：陈荣，男，教授，博导．E-mail：rchen@dl．cn
