基于多种群量子进化的区间二型模糊规则挖掘算法
Interval type–2 fuzzy rule mining algorithm based on multi-population quantum evolutionary optimization
摘要点击 97  全文点击 92  投稿时间:2017-11-28  修订日期:2018-03-17
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2018.70882
  2019,36(1):32-42
中文关键词  基于模糊规则的分类系统  量子进化算法  多种群量子编码  变尺度变异  矛盾规则重构
英文关键词  fuzzy rule-based classification system  quantum evolutionary algorithm  multi-population quantum coding  variable scale mutation  contradictory rule reconstruction
基金项目  国家自然基金项目(61102124, 61603263), 辽宁省自然科学基金项目(2015020064), 辽宁省教育厅项目(LQGD2017035)
学科分类代码  
作者单位邮编
钱小毅 沈阳工业大学 110870
张宇献 沈阳工业大学 110870
张志锋 沈阳工业大学 
王建辉 东北大学 
中文摘要
      利用智能优化算法挖掘模糊分类规则能够解决模糊前件参数和无关项的组合优化问题, 但也存在依赖初始规 则以及更新过程无指导等缺陷, 导致分类精度难以保证. 为此, 本文以二型模糊规则分类系统为框架, 采用模糊聚类得 到代表性样本并启发式的产生初始规则, 以量子等位基因形式对规则进行编码生成多初始种群, 根据基因的优良性, 以 变尺度变异操作实现等位基因的指导性进化. 在此基础上, 利用矛盾规则重构机制, 提高模糊规则分类系统的精度. 将 所提出算法与FH–GBML–IVFS–Amp算法和GAGRAD算法进行了分类精度对比, 并在不同噪声水平下, 与C4.5算 法、朴素贝叶斯分类器和BP神经网络进行分类鲁棒性比较, 实验结果表明所提出算法具有较好分类精度与鲁棒性.
英文摘要
      Employing intelligent optimization algorithm to mine fuzzy classification rule, this solves a combinatorial optimization problem on fuzzy antecedent parameters and don’t care variables. However, there are disadvantages such as the dependence of the initial rules and the lack of guidance in the updating process, which leads to it difficult to ensure classification accuracy. In this paper, the type–2 fuzzy rule-based classification system is employed as a framework, the fuzzy clustering is used to obtain the representative sample and the heuristic generation is used to generate initial fuzzy rules. The multiple initial populations are obtained by quantum alleles coding for each rule. Considering the superiority of genes, the variable scale mutation operation is used to guide the allele evolution in order to preserve the elitist population and individuals. And then, the contradictory rule is defined and the contradictory rule reconstruction is used to improve the accuracy of the fuzzy rules classification system. The classification accuracy of proposed algorithm is compared with both FH–GBML–IVFS–Amp and the GAGRAD algorithm, and classification robustness is compared with C4.5 algorithm, Naive Bayesian classifier and BP neural network at different class noise levels. The experimental results show that the classification accuracy and classification robustness of proposed algorithm are superior to compared algorithms.