Abstract:

 This paper proposes a model to fit the non-linear relation of temperature sensor based on least square
support vector machine (LS-SVM), and presents two design schemes to compensate temperature of reference point
according to the sensor’s input/output characteristics. An LS-SVM regression model is set up, which takes advantage
of the super learning capacity of LS-SVM to precisely fit the non-linear relation between temperature and voltage.
Both schemes can effectively compensate reference point temperature. The second scheme can directly fit the actual
temperature in accordance with the reference temperature and the actual measuring voltage of the sensor, simplifying
the compensation and enhancing recognition precision. Experiments show that the LS-SVM regressing method and
the proposed compensation method can fit the actual temperature and enhance measuring precision.

Key words: temperature sensor, non-linear relationship, fitting, temperature compensation, least square support, vector machine