基于PSVR的微机械陀螺温度漂移预测  

Temperature drift prediction for MEMS gyroscopes based on PSVR

作　　者：蒋蔚[1] 伊国兴[1] 曾庆双[1] 

机构地区：[1]哈尔滨工业大学空间控制与惯性技术研究中心,哈尔滨150001

出　　处：《中国惯性技术学报》2009年第3期360-365,共6页Journal of Chinese Inertial Technology

基　　金："十一五"国防预研项目(51309030102);"十一五"国防预研项目(51309030203)

摘　　要：针对中低精度微机械陀螺静态零位输出随温度漂移严重的问题,将应用于分类的近似支持向量机(PSVM)扩展到回归分析中,提出了使用近似支持向量回归机(PSVR)进行建模和预测的方法。该方法的原始优化问题基于等式约束,可采用直接法求取最优解,利用核函数可以方便地实现线性算法的非线性化,并具有良好的泛化能力。分段和连续温度测试结果表明,与常用的最小二乘支持向量回归机(LSSVR)相比,PSVR算法简单,训练速度快,尤其在大规模数据集处理上更具优势。As the static bias of low-medium accuracy MEMS gyroscopes was greatly affected by the temperature drift, the proximal support vector machines for classification were introduced into the regression analysis, and a novel modeling and prediction method based on proximal support vector regression was presented. The primitive optimization of this method was based on the equality constraints, and the optimal solution can be obtained by direct method. Using implicit mapping via the kernel function, we can easily resolve the nonlinear regression algorithms with great generalization ability. The results of temperature drift data sets from the segmented and continuous measurements demonstrate that, compared with the least squares support vector regression, the proximal support vector regression is a simple and fast training algorithm, especially for large scale data sets.

关 键 词：支持向量回归机 温度漂移 预测 微机械陀螺 

分 类 号：U666.1[交通运输工程—船舶及航道工程]