Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (10): 158-163.DOI: 10.3778/j.issn.1002-8331.1612-0379

Previous Articles     Next Articles

Study on classification of improved artificial bee colony algorithm to optimization of BP neural network

WEI Pengyu1,2,4, PAN Fucheng1,2,3, LI Shuai3   

  1. 1.Wuxi CAS Ubiquitous Information Technology R&D Center CO., LTD., Wuxi, Jiangsu 214135, China
    2.University of Chinese Academy of Sciences, Beijing 100049, China
    3.Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
    4.Research and Development Center for Internet of Things, Chinese Academy of Sciences, Wuxi, Jiangsu 214135, China
  • Online:2018-05-15 Published:2018-05-28


韦鹏宇1,2,4,潘福成1,2,3,李  帅3   

  1. 1.无锡中科泛在信息技术研发中心有限公司,江苏 无锡 214135
    2.中国科学院大学,北京 100049
    3.中国科学院 沈阳自动化研究所,沈阳 110016
    4.中国科学院 物联网研究发展中心,江苏 无锡 214135

Abstract: Due to the issues of BP neural network, which is sensitive to initial weights and easy to make the objective function into local optimum, and the disadvantages of weak local search ability and poor exploitation in standard artificial bee colony algorithm, a training method of neural network called improved artificial bee colony and back propagation is proposed. First, modify artificial bee colony algorithm inspired by the thought of differential evolution and make a more accurate description of searching behavior of onlooker bees. Second, avoid BP neural network falling into local optimum, using improved artificial bee colony to globally search the initial weights of BP neural network. Last, the datasets are tested by the new algorithm. The experiment shows compared with traditional BP neural network, the algorithm has higher classification correctness and better generalization.

Key words: BP neural network, classification, generalization, artificial bee colony

摘要: 针对BP神经网络对初始权重敏感,容易陷入局部最优,人工蜂群算法局部搜索能力和开发能力相对较弱等问题,提出一种基于改进人工蜂群和反向传播的神经网络训练方法。引进差分进化思想改进人工蜂群算法,并对跟随蜂的搜索行为进行更准确的描述。用改进的人工蜂群全局搜索神经网络的初始权重,防止神经网络陷入局部最优。用新的方法对神经网络训练进行分类。实验结果表明,该算法相对于标准的BP神经网络,有效提高了分类正确率,泛化能力较强。

关键词: BP神经网络, 分类, 泛化能力, 人工蜂群