基于协同训练与LS-SVM的集成传感器在线温度补偿  

Integrated sensor online temperature compensation based on co-training and LS-SVM

作　　者：刘继华[1] 金敏[1] 

机构地区：[1]湖南大学信息科学与工程学院,长沙410082

出　　处：《电子测量与仪器学报》2013年第11期1054-1059,共6页Journal of Electronic Measurement and Instrumentation

基　　金：国家重大科技成果转化资助项目(2012550);电子信息产业发展基金资助项目(2012407);国家自然科学基金资助项目(61374172);湖南大学青年教师成长计划资助项目

摘　　要：为利用机器学习对集成传感器实现在线补偿,使算法具有标定未知样本和更新样本集的能力,利用协同训练的方式,对最小二乘支持向量回归机进行改进,提出基于协同训练的支持向量回归算法,使用临近法对未知样本进行标定和选择,同时对新的样本空间进行剪枝,在保证反映新样本特性的前提下尽量减少对学习模型影响小的样本数量。实验证明,该算法在泛化能力不下降的情况下提高了回归精度,运用在集成传感器的在线补偿上,能降低获的成本,并取得良好的补偿效果。To use using machine learning on the online temperature compensation of the integrated sensor, the algorithm should be capable to calibrate unknown samples and update sample sets. The paper uses the co-training to improve the least squares support vector regression machine, proposes the support vector regression algorithm based on co-training, and uses the nearest neighbor method to calibrate and select the unknown samples as well as prune the new sample space, which can reflect the new features of new sample and reduce the samples which have less impact on the learning model. The experiment shows that the algorithm can guarantee the generalization capacity and improve the regression accuracy. The algorithm is used on the online temperature compensation of integrated sen- sor. It can reduce the costs of large numbers of sample calibration,and has acquired a great of achievements.

关 键 词：协同训练 最邻近点算法 支持向量机 回归 增量学习 传感器在线补偿 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]