Nonlinear Correction of Sensor based on Support Vector Machine
Denglu FANG, He ZHANG
School of Mechanical and Electrical Engineering, Southwest Petroleum University, Chengdu, CHINA
Abstract: In the process of the use of the sensor, there are various environmental factors, which lead to the error of measurement. A nonlinear correction model of the sensor based on support vector machine is proposed for the nonlinear characteristics of the influence factors and the output of the sensor. Through the establishment of training set and test set which can compare the study accuracy with correction accuracy to selections the optimal kernel function, nuclear parameters, control error and penalty factor for the support vector machine (SVM). Taking the pressure sensor as an example, the relative error of the current model compare with the BP neural network algorithm is reduced from 2.78% to 0.77%. The model significantly improves the accuracy of the sensor calibration, and has a very good application effect.
Keywords: Sensor; Support vector machine; Kernel function; Control error; Penalty factor