基于改进ELM的撑杆跳起跳过杆轨迹模型研究  

Research on trajectory model of pole vault based on improved extreme learning machine

作　　者：汪亚明[1] 李斌权 韩永华[1] 马可[2] 

机构地区：[1]浙江理工大学信息学院,杭州310018 [2]浙江理工大学体育部,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2018年第1期75-81,共7页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：浙江省自然科学基金重点项目(LZ15F020004);机械工程浙江省高校重中之重学科和浙江理工大学重点实验室优秀青年人才培养基金(ZSTUME01B17);浙江理工大学521资助项目;浙江省自然科学基金一般项目(LY17F020034)

摘　　要：目前对起跳过杆的研究仅仅专注于某一个环节,并未考虑到其他环节间的影响,为此对运动员关键部位进行标记,利用三维红外运动捕捉系统对整个起跳过杆过程进行研究。针对起跳过杆的特点,提出了一种改进的极限学习机算法。首先根据结构风险最小化理论对极限学习机进行正则化;然后利用差分进化算法的全局寻优能力对极限学习机隐层参数的选取进行优化,提高精度;最后建立起跳过杆模型。撑杆跳高的实验结果表明,该算法能有效克服极限学习机易出现过拟合的缺点,具有更好的泛化能力。仿真实验结果表明该方法有效。Recently,most of the researches on pole vault only focus on one part,but fail to take into account of the influence of other parts.So,we marked the vault with marker point at the key parts and used the three-dimensional infrared motion capture system to research the progress of pole vault.According to the characteristics of pole vault,the paper presents an improved extreme learning machine algorithm.Firstly,the theory of structural risk minimization was used to regulate extreme learning machine.Then,global optimization ability of differential evolution algorithm was adopted to optimize the selection of hidden layer parameters of extreme learning machine,which aims to improve the accuracy.After that,the model of pole vault was established.Experimental results of pole vault showed that the algorithm can efficiently overcome the overfitting shortcoming of extreme learning machine and has better generalization performance,which proves the effectiveness of the algorithm.

关 键 词：起跳过杆 极限学习机 差分进化算法 正则化 运动轨迹 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]