Computer Engineering and Applications ›› 2007, Vol. 43 ›› Issue (21): 173-176.

• 数据库与信息处理 • Previous Articles     Next Articles

Variable precision attribution reduction based on parallel artificial immune algorithm

HAO Xiao-li,XIE Ke-ming   

  1. Taiyuan Technology University,Taiyuan 030024,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2007-07-21 Published:2007-07-21
  • Contact: HAO Xiao-li

基于并行人工免疫算法的变精度属性约简

郝晓丽,谢克明   

  1. 太原理工大学 计算机学院,太原 030024
  • 通讯作者: 郝晓丽

Abstract: In order to avoid local optimization in attribution reduction with traditional artificial immune algorithm,we produce parallel artificial immune algorithm which has self-adaptive characteristic,and apply it to variable precision attribution reduction.In the algorithm,we construct similar-taxis operator,dissimilation operator and optimization operator.In similar-taxis operator,we can realize parallel searching in several sub-population.In dissimilation operator,we can exchange optimized antibodies’information among sub-population to maintain its diversity.In optimization operator,we can distribute the optimized individuals into each sub-population to realize span evolution.At the same time,the introduction uncertain measurement HVPRS as evaluation factors can adjust cross probability and mutation probability.At last,the example of fault diagnose shows that the rule sets acquired by the algorithm has better accuracy rate and coverage rate,and satisfy the requirement in fault diagnose.

Key words: parallel artificial immune algorithm, self-adaptive, variable precision, attribution reduction

摘要: 为了避免传统的人工免疫算法在属性约简时陷入局部最优解,提出了具有自适应特性的并行人工免疫算法,并且运用该算法进行粗糙集的变精度属性约简。该算法构造了趋同算子,异化算子和传优算子,利用“趋同”算子的分布性特点实现多个子种群的并行搜索,利用“异化”算子来交换种群之间优秀个体的信息,维持种群的多样性,利用“传优算子”把当前最优抗体分配到各个子群体当中,实现群体跨越式进化。在各个算子操作过程中,为了从准确度和覆盖度两方面来度量生成的规则集合的不确定性,引入了不确定量度HVPRS作为评价因子,并以此为依据,自动调整抗体的交叉概率和变异概率,使得算法不仅具有自适应的特性,而且所提取的规则集具有较高的覆盖能力和泛化能力。最后通过某发电厂发电机级故障诊断的实例,表明该算法获得的故障诊断规则集合具有较高准确度和覆盖度,满足了实际故障诊断中的要求。

关键词: 并行人工免疫算法, 自适应, 变精度, 属性约简