基于遗传算法的SFAM训练样本序列的确定  

Determination on the Sequence of SFAM Training Samples based on Genetic Algorithm

作　　者：温长吉[1] 贾鑫 迮兴业[1] 

机构地区：[1]吉林农业大学信息技术学院,吉林长春130118 [2]长春机械科学研究院有限公司,吉林长春130012

出　　处：《工程与试验》2010年第1期6-8,57,共4页Engineering and Test

摘　　要：训练样本集输入序列的确定直接关系到简单自适应谐振匹配网络(SFAM)作为分类器的执行性能,已有确定算法包括随机序列仿真算法、投票决策算法和最大-最小序列算法。基于遗传算法作为一种全局搜索算法的特点,本文提出用遗传算法实现SFAM训练样本集最优输入序列确定的方法,并以加州大学Irvine分校机器学习数据库作为实验样本库,实验结果表明,该算法比投票决策算法和最大-最小序列算法对提升SFAM网络分类器的分类精度和降低训练时间更为有效。The presentation order of training patterns to a simplified fuzzy ARTMAP（SFAM）neural network affects the classification performance.The current algorithm includes training patterns presented in random order,voting strategy and max-min ordering method.Based on genetic algorithm as a global search algorithm,in this paper,a method that uses genetic algorithm（GA）to select the presentation order of SFAM training patterns is proposed.In addition,three datasets from UCI repository were used as test sample datasets.The performances of the proposed method have been compared with the performances of the voting strategy and the min max method,and the experimental results show that the proposed method is more effective in enhancing the classification accuracy and reducing training time of SFAM.

关 键 词：自适应谐振匹配网络 遗传算法 样本序列 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]