摘要
结合最小熵原理法(MEPA)的数据离散功能,粗糙集理论(RST)的数据分析和容错能力,以及朴素贝叶斯网络分类器(NBNC)的并行推理能力,采用串行集成思想,提出了一种基于MEPA-RST-NBNC的复杂设备智能故障诊断方法。首先利用MEPA实现连续条件属性的离散化,形成离散化诊断决策表;然后利用RST分辨矩阵实现故障特征的简化,并采用最大聚类比原则提取出最佳约简;最后根据约简诊断决策表建立NBNC模型来实现高效快速的诊断推理。故障诊断实例表明该方法不仅克服了RST诊断法的规则搜索和临界误判问题,而且避免了NBNC诊断法的维数灾难问题,具有较强的工程实用性。
According to serial ensemble strategy, and considering the ability of minimize entropy principle approach （MEPA） for discretizing data, the ability of rough set theory （RST） for analyzing and reducing data and the ability of naive Bayesian network classifier （NBNC） for parallel inference, a novel intelligent fanlt diagnosis method based on MEPA-RST-NBNC is proposed. Firstly, the continuous attributes in original diagnostic decision table are discretized using the MEPA-based discretization method, and consequently the discrete diagnostic decision table is obtained. Then, the redundancy conditional attributes are reduced utilizing the RST discernibility matrix, and further, the optimal attribute reduction is established using the maximum cluster ratio principle. Finally, the NBNC model is built based on the reduction diagnosis decision table to carry out fault diagnosis. The results of a fault diagnosis example show that the proposed method not only can overcome the rule search and critical misjudgment problems of RST-based fault diagnosis method, but also can avoid the dimensional disaster problem in NBNC-based diagnosis method, so it has strong engineering practicality.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2008年第12期2480-2485,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(60572173)
航天创新基金(2004CH01001)
西北工业大学青年科技创新基金(M016217)资助项目
关键词
最小熵原理方法
粗糙集
朴素贝叶斯网络分类器
故障诊断
minimize entropy principle approach
rough set theory
naive Bayesian network classifier
fault diagnosis
