An Improved LSSVR-Based Nonlinear Calibration for Thermocouple

Abstract

A new approach to nonlinear calibration of thermocouple based on a improved least squares support vector regression machine (LS-SVR) is proposed. Firstly, the response of compensator based on the principle of inverse model is expressed in terms of thermocouplepsilas output by a power series. Therefore, the nonlinear calibration of thermocouple is transformed to the identification problem of compensator model. Then, aiming at the calibration data set with n data points and m features and n>>m, Sherman-Morrison-Woodbury (SMW) transformation is introduced, through which solving a LSSVR only involves inverting an m dimensional matrix instead of n dimensional one. Lastly, the data of platinum-rhodium 30-platinum-rhodium 6 thermocouple(B) are used to test and the experiment results demonstrate that the computational complexity of improved LSSVR is independent of the sample size n, and the efficiency of which is superior. Thus this compensation technique provides faster calibration on a large sample condition.