基于群集正反向回溯改进人工生态系统优化算法的ELM超参优选
作者:
作者单位:

辽宁工程技术大学

作者简介:

通讯作者:

中图分类号:

TP391; TP301.6

基金项目:

辽宁省科技厅博士科研启动基金项目(2019-BS-118);辽宁省教育厅基金项目(LJ2019JL017)


Super parameter optimization of ELM based on improved artificial ecosystem-based optimization with crowding forward-backward and backtracking
Author:
Affiliation:

Liaoning Technical University

Fund Project:

  • 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:

    为提高人工生态系统优化算法(AEO)的全局勘探和局部极值规避性能,提出一种基于群集正反向回溯改进人工生态系统优化算法(IAEO),并用于高维优化. 首先,启发于自然生态系统中消费者数量因捕食关系而发生的动态波动变化现象,借鉴上下级消费者间的数量正反向调控机理而设计并模型化表达群集正反向引导机制,以增强AEO算法在解空间的全局勘探性能;其次,构造局部回溯开采策略,其继承种群历史最优解信息对分解者(最优个体)的局部微小邻域进行动态再挖掘,并进一步引导种群个体进化,以改善算法的局部极值规避性能. IAEO与7种新近智能算法在100维度上进行对比,同时与AEO算法在高维下进行分析,数值实验表明IAEO算法具有较好的全局探索和局部开采性能、良好的算法局部极值规避性和鲁棒性,以及在高维情形下的较好适用性(200D-1000D). 最后,将IAEO算法应用于极限学习机(ELM)神经网络的超参优选中,提出一种新颖的基于IAEO算法的ELM参数优化模型(IAEO_ELM),并与其它ELM优化模型执行分类任务,仿真结果表明IAEO_ELM具有优越的分类性能,能有效提升算法的分类预测精度.

    Abstract:

    To improve the global exploration and local extremum avoidance performance of artificial ecosystem optimization algorithm (AEO), an improved artificial ecosystem-based optimization algorithm (IAEO) based on crowding forward-backward and backtracking was proposed, which is also used for high-dimensional optimization. At First, inspired by the phenomenon that the number of consumers in a natural ecosystem fluctuates dynamically due to predator-prey relationship, to design and model the crowding forward-backward guidance mechanism by referring to the forward and reverse regulation mechanism of the number of consumers between the upper and lower levels, so as to enhance the global exploration performance of AEO in the solution space. What’s more, a local backtracking exploitation strategy is constructed, which dynamic remining local small neighborhood of decomposers (optimal individuals) by inheriting populations historical optimal solution information, and further guides the individuals evolution to improve the local extremum avoidance performance of the algorithm. IAEO is compared with seven intelligent algorithms in 100 dimensions, and compared with AEO in high-dimensional, numerical experiments show that IAEO has better global exploration and local exploitation performance, good local extremum avoidance performance of the algorithm and robustness, and also has good applicability in high-dimensional situations(200D-1000D). Finally, IAEO is applied to the optimization of extreme learning machine (ELM) neural network, and a novel ELM parameter optimization model based on IAEO (IAEO_ELM) is proposed. Then perform classification tasks with other ELM optimization models, the simulation results show that IAEO_ELM has superior classification performance and can effectively improve the classification prediction accuracy of the algorithm.

    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:2021-09-17
  • 最后修改日期:2022-01-03
  • 录用日期:2022-01-11
  • 在线发布日期: 2022-03-01
  • 出版日期: