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

It is an efficient approach of nonlinear correction for sensors by fitting curves to data, but it is difficult to fit curve with a single model for the thermocouple vacuum sensor, since its characteristic is very complex. A novel method of curve fitting based on multiple support vector regression (MSVR) is proposed. The sample space are divided into several sub-spaces, nonlinear mapping is established in each sub-space by using support vector regressions (SVR), each SVR may has parameters different from the others to approximate the sensor characteristic in corresponding partial space, combining the outputs of all the SVRs the complete characteristic of the sensor is achieved. The proposed approach was applied to fit the characteristic curve for the thermocouple vacuum sensor, the experiment results demonstrate the effectiveness and practicability of the proposed method.