摘要
为了逼近光纤陀螺(FOG)温度漂移的复杂非线性关系,提出一种基于有界整体经验模态分解(BEEMD)和极限学习机(ELM)的多尺度集成建模方法(SE-BEEMD-ELM)。采用样本熵(SE)分析BEEMD分解得到的本征模态函数(IMF)序列,根据SE值变化趋势和大小得到漂移数据的多个尺度分量。分别以温度梯度变化和单一尺度分量训练ELM子模型,累加生成的多个子模型得到FOG温度漂移的集成模型。实验结果表明,基于SE-BEEMD-ELM的多尺度集成建模方法,建模精度较基于BEEMD-BP以及BEEMD-ELM的单一模型提高2个量级。
To approximate the complicated nonlinear relationship between temperature and FOG drift,a novel multiscale modeling method is presented based on bounded ensemble empirical mode decomposition( BEEMD) and extreme learning machine( ELM)( designated as SE-BEEMD-ELM).Sample entropy( SE) is utilized to analyze all the intrinsic mode functions( IMF) produced by BEEMD,obtaining multiple sub-components of FOG drift by accumulating the IMFs according to the variation trend of SE values.Single ELM model is used time-dependent temperature gradients and sub-component as the input variables.Finally,all of the single models are combined to produce an ensemble model.Experiment result shows that compared with BEEMD-BP and BEEMD-ELM,which are based on the single model,the modeling accuracy of SE-BEEMD-ELM is improved by two orders.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2015年第2期304-309,共6页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(51375087
50975049)
中央高校基本科研业务费专项资金(3222004905)
江苏省普通高校研究生科研创新计划(KYLX_0106)资助项目
关键词
光纤陀螺
经验模态分解
极限学习机
温度漂移
fiber optic gyroscope
empirical mode decomposition
extreme learning machine
temperature drift