Abstract: Aiming at feature selection and parameter optimization problem of classifier based on Support Vector Machine(SVM), one kind of combination optimization for feature selection and parameter is presented based on chaotic genetic algorithm of mutative scale. The method of chromosome coding and decoding is presented. Initialization population is produced by using chaos ergodicity. A modified crossover operator and dynamic reducing searching space is used for further optimization. This method is applied to fault classifier combination optimization of shortwave-control-equipment. Experimental results assess effectiveness on finding optimal solution of the proposed approach.

Key words: Support Vector Machine(SVM), chaotic genetic algorithm, feature selection, parameter optimization, fault diagnosis, mutative scale

摘要： 为解决支持向量机(SVM)分类器的样本特征选择和参数优化问题，提出一种将特征选择和参数选择进行联合优化的方法。基于变尺度的混沌遗传算法，联合优化染色体编、译码，利用混沌的遍历性产生初始种群，改进遗传算法中的交叉算子，动态缩减寻优区间。将该方法应用于短波通信控制器的诊断分类器中，以实现分类器特征子集选取和参数的联合优化，结果表明该方法具有较强的寻优能力。

关键词: 支持向量机, 混沌遗传算法, 特征选取, 参数优化, 故障诊断, 变尺度

CLC Number: 

• TP273