文献类型：期刊文献

中文题名：超限学习机在磁罗盘非线性误差补偿中的应用

英文题名：Application of extreme learning machine in the nonlinear error compensation of magnetic compass

作者：刘艳霞[1];方建军[1];张晓娟[2];孙建[1]

第一作者：刘艳霞

通讯作者：Fang, Jianjun

机构：[1]北京联合大学自动化学院;[2]北京科技大学检测与控制系

第一机构：北京联合大学城市轨道交通与物流学院

年份：2015

卷号：36

期号：9

起止页码：1921-1927

中文期刊名：仪器仪表学报

外文期刊名：Chinese Journal of Scientific Instrument

收录：CSTPCD;;EI(收录号：20154501501913);Scopus(收录号：2-s2.0-84945922019);北大核心:【北大核心2014】；CSCD:【CSCD2015_2016】；

基金：国家自然科学基金(61273082);北京市自然科学基金(4142018);北京市教委(KM201511417004);北京联合大学启明星大学生科技创新(201511417011)项目资助

语种：中文

中文关键词：超限学习机;神经网络;非线性误差模型;磁罗盘

外文关键词：extreme learning machine; neural network; nonlinear error model; magnetic compass

摘要：针对磁阻式传感器组成的磁罗盘中不容忽视的非线性误差,建立了隐式非线性误差模型,引入机器学习中的超限学习机算法对非线性误差模型进行训练。利用训练好的误差模型对航向角测量误差进行补偿,误差由补偿前的±3°下降到±0.2°,均方根误差为0.1°。任意选取的训练集、测试集和重复实验证明超限学习算法具有很好的泛化性和鲁棒性,而且训练速度极快,是传统BP神经网络的上千倍。
Aiming at the nonlinear error that cannot be ignored in the magnetic compass composed of magnetic resistance sensor, an im- plicit nonlinear error model is established. The extreme learning machine algorithm in machine learning is introduced to train the nonlinear error model. The trained error model is used to compensate the heading measurement error, which is decreased from ±3° before compensation to ±0.2° after compensation, and the root mean square error is 0.1°. The training result for randomly chosen training and testing sets and the repeated experiment result show that the extreme learning machine algorithm has good generalization and robustness, and extremely fast training speed, which is thousands of times of that for traditional BP neural network.