基于粒计算的知识获取方法研究及其应用

摘要

Aiming at the difficulties of complex knowledge reduction process and low knowledge acquisition efficiency of rotating machinery fault decision table,through the analysis of the nature of knowledge discovery and according to the requirements of knowledge reduction of the decision table,a fault attributes reduction method based on granular computing was proposed.This method divided the universe into equivalence classes with decision attribute,and then added attributes for meeting conditions in reduction sets to obtain minimal attribute reduction sets of different faults by condition information entropy and starting at the equivalent classes of decision attribute.The attributes reduction method was applied to the rotor fault decision table,and then the diagnosis rules for five kinds of faults were obtained.The evaluating results show that the method can achieve more effective performance for rotor fault decision table.%针对旋转机械故障决策表知识约简过程复杂、知识获取效率低下的问题,通过对知识发现问题本质的分析,结合决策表知识约简的要求,提出了一种基于粒计算的故障特征属性约简方法。该方法先用决策属性划分整个论域,然后从决策属性的各个等价类出发,以条件熵为启发信息,通过向约简集中逐渐添加满足条件的属性来获得各种故障的最小属性约简。将此方法应用于转子故障决策表,得到了5种典型故障的诊断知识规则。知识规则支持度和置信度的评价结果表明,该方法对转子故障决策表具有较好的约简效果。