基于改进极限学习机的MEMS陀螺随机误差补偿方法研究  

Research on Compensation Method of MEMS Gyroscope Random Error Based on Improved Extreme Learning Machine

作　　者：杨辉 杨川 姜湖海[1] 王晶[1] 毛锐 YANG Hui;YANG Chuan;JIANG Huhai;WANG Jing;MARui(Southwest Institute of Technical Physics;The Second Military Representative Office of Chongqing Military Representative Bureau of Army Equipment Department in Chengdu,Chengdu Sichuan 610041,China)

机构地区：[1]西南技术物理研究所,四川成都610041 [2]陆军装备部驻重庆地区军代局驻成都地区第二军代室,四川成都610041

出　　处：《传感技术学报》2021年第2期244-248,共5页Chinese Journal of Sensors and Actuators

摘　　要：针对微机械(Micro Electro Mechanical System,MEMS)陀螺随机误差较大、精度不高的问题,提出了一种基于改进激活函数的极限学习机MEMS陀螺建模的方法。首先采集转台和MEMS陀螺数据作为训练集数据,随后利用粒子群优化算法确定隐含层神经元数目、陡度因子、位置参数、映射区间因子等参数,最后采用改进激活函数的极限学习机进行建模。在MEMS陀螺单体转台试验结果表明,该方法较原极限学习机、卡尔曼滤波、小波滤波等三种方法的均方误差和平均绝对误差指标更优,其中均方误差可提高一个数量级。在导引头稳定平台闭环测试结果表明,该模型补偿后Allan方差指标均得到明显改善,能满足系统使用要求。Aiming at the large random errors and low precision of Micro Electro Mechanical System(MEMS)gyroscope,an Extreme Learning Machine(ELM)modeling method based on fixed activation function is proposed in this paper.Firstly,the training database of turntable and MEMS gyroscope are collected,then the number of hidden layer nodes,steep degree,position and mapping scope is optimized by using particle swarm optimization(PSO)simultaneously.The finally model is building by ELM with fixed activation function.The results of in turntable experiments show that method is better than others method like the original ELM,Kalman filter and wavelet filter by the Mean Squared Error(MSE)and Mean Absolute Error(MAE),and raise on order of magnitude of MSE.The results of in seeker system experiments show that effectiveness of the method based on Allan variance analysis,and satisfy the seeker system requirements.

关 键 词：MEMS陀螺 极限学习机 导引头 ALLAN方差 

分 类 号：U666.1[交通运输工程—船舶及航道工程]