基于SAPSO算法的RBF神经网络设计
作者:
作者单位:

河南理工大学

作者简介:

通讯作者:

中图分类号:

TP183

基金项目:

国家自然科学基金(61703145),河南省高校科技创新团队(20IRTSTHN019)


Design of RBF Neural Network Based on SAPSO Algorithm
Author:
Affiliation:

Henan Polytech University

Fund Project:

National Natural Science Foundation of China (61703145),Science and technology innovation team of henan university(201RTSTHN019)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对径向基神经网络结构和参数的动态优化问题,本文提出一种基于敏感度分析和粒子群优化的RBF 神经网络(SAPSO-RBF) 优化算法. 算法通过随机初始化各粒子信息数,基于粒子敏感度分析,对算法学习阶段 粒子信息进行增加和删减,确定算法第一次收敛时网络结构大小;算法达到收敛后,对最优粒子进行敏感度分 析,删除冗余信息,使算法重新发散;根据算法发散和收敛次数提出一种惯性权重更新方法,使算法在解空间 内进行多次发散和收敛,增强算法搜索能力同时减小网络结构,并给出SAPSO 算法的收敛性证明. 仿真实验研 究表明,SAPSO-RBF 算法具有良好的自组织能力,相较于一些其他自组织RBF 神经网络优化算法,在网络结 构紧凑度和精度等方面有较大提升.

    Abstract:

    Aiming at the dynamic optimization of structure and parameters of radial basis function (RBF) neural network, an optimization algorithm based on sensitivity analysis (SA) and particle swarm optimization (PSO) for RBF neural network (SAPSO-RBF) is proposed. Firstly, the number of particle information is randomly initialized, and particle information is added and deleted by the sensitivity analysis in the learning phase, and the size of the network structure of the algorithm first converges is determined. Then, After the algorithm reaches convergence, it analyzes the sensitivity of the optimal particles, deletes the redundant information, and makes the algorithm re-divergent; an inertia weight update method is proposed to make the algorithm perform multiple divergence and convergence in the solution space, which enhances algorithm search ability while reduces network structure, the convergence of SAPSO algorithm is proved in this paper. Finally, the results of experiments indicate that the proposed SAPSO-RBF algorithm has the good self-organizing ability, it has greatly improved the network structure compactness and accuracy compared with some other existing methods.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-02-23
  • 最后修改日期:2020-06-11
  • 录用日期:2020-06-12
  • 在线发布日期:
  • 出版日期: