摘要
将极限学习机引入卫星钟差预报,对比Sigmoidal、Sine和Hardlim三种激励函数对预报精度的影响,并与传统灰色系统模型和径向基函数神经网络进行比较。结果表明,极限学习机的预报精度优于另外两种模型,其学习速度也快于径向基函数神经网络,且基于Sigmoidal的激励函数最适合于钟差预报。
Extreme Learning Machine (ELM) is employed for predicting satellite clock error. The impact of activation functions on prediction accuracy using ELM is analyzed, including Sigmoidal, Sine and Hardlim functions, and ELM model is compared with the grey system model and radial basis function (RBF) neural network(NN) model. The results show that prediction precision of ELM algorithm is best, and can learn faster than RBF neural network. Moreover, the Sigmoidal activation function is best for clock error prediction.
出处
《大地测量与地球动力学》
CSCD
北大核心
2013年第5期53-57,共5页
Journal of Geodesy and Geodynamics
基金
国家自然科学基金(10573019)
关键词
极限学习机
激励函数
径向基函数神经网络
卫星钟差
钟差预报
Extreme Learning Machine (ELM)
activation function
radial basis function neural network
satelliteclock error
clock error prediction