摘要
利用神经网络进行光电轴角编码器的误差补偿,是一种新的编码器误差修正技术。在利用径向基函数网络(RBF)建立非线性模型对编码器的长周期误差进行修正时,实验数据中的随机噪声会造成过拟合现象的产生,影响模型的补偿效果。针对RBF网络的过拟合现象,分别采用滑动平均法与异常点剔除法对实验数据进行预处理以减小随机噪声的影响。实验表明,使用径向基神经网络进行误差补差可以将编码器的系统精度提高至3倍以上,在此基础上对建模数据进行预处理,可以进一步优化网络性能,使得编码器的系统精度在原有基础上得到进一步的改善,验证了方法的有效性。
A method to apply the RBF to the optical encoder system for error correction and compensation. For Radial Basis Function(RBF)network modelling, random error contained in the experimental data can cause the over-fitting phenomena and then influenced the model’s precision. Two strategies for pretreatment to original data were introduced to cope with the defect, that were data’s moving average and outlier data’s diagnosis and deletion. The results show that models set up with pretreated data have better perfor...
出处
《红外与激光工程》
EI
CSCD
北大核心
2008年第S1期87-89,共3页
Infrared and Laser Engineering
基金
国家自然科学基金资助项目(60574089)
