基于多支持向量回归机的热偶规真空传感器非线性校正研究  

Nonlinear Correction for Thermocouple Vacuum Sensor Based on Multiple Support Vector Regressions

作　　者：高飞燕[1] 唐耀庚[1] 刘华[1] 

机构地区：[1]南华大学电气工程学院,湖南衡阳421001

出　　处：《测试技术学报》2010年第1期29-33,共5页Journal of Test and Measurement Technology

摘　　要：曲线拟合是传感器非线性校正的重要方法,但热偶规真空传感器输出特性具有明显的分段性且各段特性差异大,采用单一模型难以取得满意的拟合结果.本文提出基于多支持向量回归机(MSVR)的曲线拟合方法,将输入样本空间分割成多个子空间,为每一子空间建立一个支持向量回归机(SVR)来映射该局部空间的非线性关系,合成各SVR的输出即为传感器全量程输出特性.用所提出的方法拟合热偶规真空传感器输出特性,仿真结果表明了该方法的有效性.Curves fitting is an important approach of nonlinear correction for sensor complex output characteristics, it is difficult for a thermocouple vacuum sensor to obtain s. But due to high-precision fitting with a single model. Here we propose a curve fitting method based on multiple support vector regression （MSVR）. The input samples are divided into some sub-spaces, and mapping relation is set up for each sub-space by using a support vector regression （SVR） which has parameters different from others. Each SVR approximates the sensor characteristic of the corresponding area. The sum of all SVR output is the full-scale characteristic output of the sensor. The method was applied to fit the output characteristic curve for the thermocouple vacuum sensor. The simulation results show that this approach is valid.

关 键 词：多支持向量回归机 样本空间分割 拟合 非线性校正 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]