基于GA-DBN模型的分类方法研究

Research on classification method based on GA-DBN model

本文采用多层受限玻尔兹曼机和BP神经网络构成的深度置信网络作为分类模型,并采用遗传算法对深度置信网络中的受限玻尔兹曼机的参数进行逐层寻优,从而构建了GA-DBN分类模型。通过与随机森林、人工神经网络、支持向量机和XGBoost分类模型进行对比实验发现,构建的GA-DBN模型在分类准确度和F_(1)值上均有较好的表现。为验证遗传算法对DBN模型的优化效果,将人工蜂群算法优化的DBN模型即ABC-DBN模型与GA-DBN模型进行对比实验发现,GA-DBN模型在准确率、F_(1)值和运行时间上均表现出良好的效果,说明遗传算法对DBN模型的优化效果更突出。

In this paper,Deep Belief Networks(DBN)composed of Restricted Boltzmann Machine(RBM)and BP neural network were adopted as classification model.The GA-DBN classification model is constructed by using Genetic Algorithm(GA)to optimize the parameters of constrained Boltzmann machine in deep confidence network layer by layer.By comparing with random forest,artificial neural network,support vector machine and XGBoost classification model,it is found that the classification accuracy and F_(1) value of GA-DBN model are better.In order to verify the optimization effect of genetic algorithm on DBN model,the DBN model optimized by Artificial Bee Colony algorithm(ABC),namely ABC-DBN model and GA-DBN model,was compared and tested.GA-DBN model shows good results in accuracy,F_(1) value and running time,which verifies that genetic algorithm is more effective in optimizing DBN model.