摘要
MEMS陀螺的零偏随温度呈非线性变化,同时含有较大的随机噪声。针对传统的多项式模型难以精确表达零偏随温度变化的问题,提出了一种基于灰色模型和RBF神经网络的MEMS陀螺温度补偿方法:首先用灰色模型对数据进行预处理,以减小原始数据的噪声;然后用降噪后的样本数据对RBF神经网络进行训练。在相同的训练次数下训练误差可减小一个数量级。验证试验结果表明,采用该模型补偿后的陀螺零偏误差较传统的多项式模型减小一个数量级,较未经预处理的RBF神经网络减小2/3。
The zero bias of MEMS gyroscope exhibits nonlinear change with varying temperature and contains significant stochastic error.In view that traditional polynomial method could not accurately establish the relationship between the temperature and the zero bias,this paper puts forward a hybrid model based on grey model theory and RBF neural network.First,it pre-processes the gyro output using the grey model to reduce the noise.Then it uses the processed sample data to train the RBF network,so the training errors are reduced by one magnitude within the same training times.The verification experiments indicate that compared with traditional polynomial model and untreated RBF neural network,the compensated gyro zero bias error is reduced by one order of magnitude and 2/3 respectively.
出处
《中国惯性技术学报》
EI
CSCD
北大核心
2010年第6期742-746,共5页
Journal of Chinese Inertial Technology
基金
天津市科技支撑重点项目(08ZCKFGX04000)
