基于免疫文化算法的故障特征选择方法  

Immune cultural algorithm for fault feature selection

作　　者：宋辰[1] 黄海燕[1] 

机构地区：[1]华东理工大学信息科学与工程学院,上海200237

出　　处：《计算机应用研究》2012年第11期4162-4164,4175,共4页Application Research of Computers

摘　　要：提出了一种新的文化算法,基于免疫克隆选择原理改进了文化算法的种群空间,同时设计了一种新的历史知识及其影响函数。为了去除工业中故障诊断过程中的冗余变量,实现数据降维,提高故障诊断性能,将该免疫文化算法应用到故障特征选择当中,提出了一种封装式的特征选择方法。该方法利用抗体种群进行全局搜索,通过文化算法的信念空间保留历代最优个体,并对UCI数据集的高维数据进行特征子集选择。将该方法应用到TE过程故障诊断中,结果表明,相比于直接使用高维数据进行故障诊断,该算法有效降低了特征空间的维数,提高了分类精度。This paper proposed a new cultural algorithm.It applied the principle of immune clone selection as population space of cultural algorithm,meanwhile designed a new knowledge of history and its influence function.This paper proposed a wrapper feature selection algorithm based on immune cultural algorithm to remove redundant variable in the industrial process fault diagnosis process,reduced the data dimension and improved the performance of fault diagnosis.The method used antibody populations to global search and retained the best individual through the belief space of cultural algorithms.And it used high dimensional data of UCI data sets to test this method.The results show that compared with diagnose fault directly,the algorithm effectively reduces the dimension of the feature space and improves the classification accuracy in fault diagnosis of Tennessee Eastman（TE） process.

关 键 词：故障诊断 特征选择 免疫克隆选择 文化算法 TE过程 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]