一种基于超限学习机的电子磁罗盘非线性误差补偿方法

A Nonlinear Error Compensation Method for Electronic Magnetic Compass Based on Extreme Learning Machine

针对磁罗盘传感器非线性校正中现有方法的不足,提出采用小波函数和双曲正弦函数作为超限学习机(ELM)的激活函数,并将此改进超限学习机用于磁罗盘的校正。同时,阐述了传感器的非线性校正原理,磁罗盘航向误差模型及改进超限学习机的实现过程,并分别采用BP神经网络法和传统ELM对磁罗盘进行非线性校正。实验结果表明,改进ELM算法补偿后最大误差为0.103°,均方根误差为0.0596°),优于BP神经网络算法(补偿后最大误差为0.5°,均方根误差为0.1805°)和传统ELM神经网络(补偿后最大误差为0.21°,均方根误差为0.1056°)。

In order to overcome the disadvantages of commonly methods used in nonlinear correction of electronic magnetic compass（EMC）, an approach that applies wavelet function and hyperbolic sine function as the activation function of extreme learning machine（ELM） is presented, then the improved ELM is used for the correction of EMC. Meanwhile, the nonlinear correction principle of sensors, the realization processes of EMC heading error model and improved extreme learning machine（IELM） were expounded, and the nonlinear correction of EMC was adjusted with BP neural network and the method of traditional ELM, respectively. The experiment results show the IELM＇s maximum error is 0. 103° after compensation and root mean square error is0. 0596°, which is better than that of BP neural network whose maximum error after compensaton is 0. 5° and root mean square error is 0. 1805°, and traditional ELM whose maximum error after compensation is 0.21° and root mean square error is 0. 1056°.