基于遗传小波神经网络的MEMS陀螺误差建模  

MEMS gyroscope’s error modeling based on wavelet neural network of genetic algorithms

作　　者：卢海曦[1] 夏敦柱[1] 周百令[1] 

机构地区：[1]东南大学仪器科学与工程学院,南京210096

出　　处：《中国惯性技术学报》2008年第2期216-219,共4页Journal of Chinese Inertial Technology

基　　金：国家863高技术项目(2002AA812038)

摘　　要：研究了一种基于遗传算法改进的小波神经网络。该方法采用小波神经网络为主要逼近手段,并通过遗传算法优化网络的关键参数。由于小波神经网络兼容了神经网络的自学习特性和小波分析的时—频局部性,而遗传算法具有较强的全局搜索能力,二者结合形成的遗传小波神经网络因此具有较高的逼近与容错能力,从而弥补了传统的方差建模方法的不足。将该算法应用到某型MEMS陀螺仪的随机误差建模中,结果表明:遗传小波网络对原始信号的逼近精度误差在1.75%以内,较之传统的方差建模方法有了显著的提高,这一精度基本上可以满足MEMS陀螺工程化应用的要求。A Wavelet Neural Network improved by Genetic Algorithms（GA） was presented. The method employs wavelet neural network as main approaching tool, which optimize the parameters by GA. Since the Wavelet neural network combines the self-learning ability of neural network with the time-frequency localization of the wavelet analysis, while the Genetic Algorithms（GA） employs the characteristic of global optimum searching, the wavelet neural network improved by GA holds strong approaching ability, which overcome the disadvantage of traditional modeling method. Its application in modeling a MEMS gyro＇s random drift shows that the model has an identification error within 1.75%. This precision can reach the requirements in engineering.

关 键 词：MEMS陀螺 随机误差 遗传小波神经网络 建模 

分 类 号：U666.1[交通运输工程—船舶及航道工程]