基于递阶遗传算法和BP网络的模式分类方法  

HGA-BP-Based Pattern Classification Method

作　　者：周辉仁[1] 郑丕谔[1] 牛犇[1] 宗蕴[2] 

机构地区：[1]天津大学系统工程研究所,天津300072 [2]山东大学能源与动力学院,济南250100

出　　处：《系统仿真学报》2009年第8期2243-2247,共5页Journal of System Simulation

摘　　要：提出一种基于递阶遗传算法和BP神经网络的模式分类模型。现有的BP网络模式分类训练方法大都只能训练BP网络的权重,网络的结构得预先用某种方法确定。利用巧妙设计的递阶遗传算法能够把网络的结构和权重同时通过训练确定。以模式分类数据库中的数据进行训练和测试,并与其它模式分类模型相比较,结果证明该模型更优,分类精确度更令人满意的,所提出的方法是可行的。A pattern classification model based on hierarchical genetic algorithm and BP neural network was proposed. Different from the existing BP training method resulting only in determination of connection weights, a well-designed hierarchical genetic algorithm was used to train BP neural network with both connection weights andnumbers of neurons in a hidden layer determined at the same time. The model was then used to classify the data of Iris and Wine. It is shown that the model based on hierarchical genetic algorithm and BP neural network is simple and effective.

关 键 词：递阶遗传算法 训练方法 BP神经网络 权重与结构 模式分类 

分 类 号：O235[理学—运筹学与控制论]