基于极限学习机的三维图形重构虚拟仿真实验研究

摘要

In view of the limitations of over fitting and small generalization ability caused by extreme learning machine (ELM),it uses moving least square extreme learning machine (MLS-ELM) and regularized extreme learning machine (R-ELM) optimization algorithm to balance structure risk and empirical risk caused by extreme learning machine (ELM).The design improves the generalization ability of the ELM.It uses MLS-ELM and R-ELM to conduct the virtual simulation experiment about the reconstruction of 3D graphics of the Mexican hat function.This experiment shows that MLS-ELM and R-ELM can effectively reduce the reconstruction error.This simulation experiment can be used in the neural network and Matlab virtual simulation teaching,and plays a positive role in improving the students' autonomous learning ability and the programming ability.%针对极限学习机容易导致过拟合、泛化能力小等局限性,采用移动加权极限学习机和正则极限学习机优化算法,平衡原始极限学习机存在的结构风险和经验风险,提高极限学习机的泛化能力;并用该算法对墨西哥帽子函数进行三维重构虚拟仿真实验.实验表明,这两种算法能够有效的降低重构误差,提高算法的泛化能力.该仿真实验可用于神经网络及Matlab虚拟仿真实验教学,对提高学生自主学习能力,编程与调试能力起到积极作用.