摘要
轴系故障一直是困扰电力生产的主要问题,基于专家系统的轴系故障诊断方法存在知识获取困难、诊断精度不高的问题。本文提出了基于归纳学习的分类树构造方法,实现从运行数据中提取诊断规则,分析了基于分类归纳的诊断规则抽取方法对噪声的容忍能力,在不同噪声类型和噪声规模下,分析发现属性约简和剪枝技术都能有效提高诊断规则的泛化性能。本文进一步提出了将数值型规则转化为模糊规则,提高诊断规则的可理解性和推广能力,为实现故障诊断系统的自学习提供了可行的途径。
Shaft faults is one of the main problems which cause turbo-generator breaks. Fault diagnosis based on expert systems is implemented for shafts in some power plant. Automatic learning is the core issue in applications of expert system. In this paper, an induction algoritlun for classification trees is introduced. The algorithm is able to construct a classification tree from training da'ta and transform the tree to a set of rules, which are widely used in expert systems. Then generalization of the induced rule set is analyzed based on minimal description length. We introduce support and confidence as measures of rule strength and give a pruning method. The experimental results show both attribute reduction and pruning methods will improve the recognition performance.
出处
《节能技术》
CAS
2005年第4期331-334,347,共5页
Energy Conservation Technology
关键词
轴系故障诊断
归纳学习
分类树
规则抽取
fault diagnosis
induction learning
classification tree
rule extraction