基于灵敏度分析法的ELM剪枝算法.pdfVIP

第29 卷 第6 期 控 制 与 决 策 2014 年 6 月 Vol. 29 No. 6 Control and Decision Jun. 2014 文章编号: 1001-0920 (2014) 06-1003-06 DOI: 10.13195/j.kzyjc.2013.0306 基于灵敏度分析法的ELM 剪枝算法 1,2 1 1 李凡军 , 韩红桂 , 乔俊飞 (1. 北京工业大学电子信息与控制工程学院，北京100124；2. 济南大学数学科学学院，济南250022) 摘 要: 针对极端学习机(ELM) 网络结构设计问题, 提出基于灵敏度分析法的ELM 剪枝算法. 利用隐含层节点输出 和相对应的输出层权值向量, 定义学习残差对于隐含层节点的灵敏度和网络规模适应度, 根据灵敏度大小判断隐含 层节点的重要性, 利用网络规模适应度确定隐含层节点个数, 删除重要性较低的节点. 仿真结果表明, 所提出的算法 能够较为准确地确定与学习样本相匹配的网络规模, 解决了ELM 网络结构设计问题. 关键词: 前馈神经网络；极端学习机；灵敏度分析；剪枝算法 中图分类号: TP273 文献标志码: A Pruning algorithm for extreme learning machine based on sensitivity analysis 1,2 1 1 LI Fan-jun , HAN Hong-gui , QIAO Jun-fei (1. College of Electronic Information and Control Engineering，Beijing University of Technology ，Beijing 100124, China ；2. School of Mathematical Science，Ji’nan University ，Ji’nan 250022 ，China ．Correspondent ：QIAO Jun- fei ，E-mail ：junfeiq@bjut.edu.cn) Abstract: In order to design the structure of extreme learning machine(ELM), a pruning algorithm is proposed by using the sensitivity analysis method. The residual error’s sensitivities to the hidden nodes are defined by their outputs and weight vectors connecting to the output layer. The model scale adaptability is calculated and the hidden nodes are sorted by using the defined sensitivities. Then, the number of requisite hidden nodes is estimated by the model scale adaptability. The redundant nodes with smaller sensitivities are removed from the existent network. The simulation results show that the proposed approach can construct