Hysteresis and nonlinearity compensation of relative humidity sensor using support vector machines

Based on support vector machine (SVM) technique, a novel method for hysteresis and nonlinearity compensation of a relative humidity sensor has been investigated. The compensation method consists of a two-stage procedure. First, a SVM is used to perform the hysteresis error compensation of the sensor. Then, a second SVM is utilized to compensate the nonlinearity error in the response characteristics and to estimate the applied actual relative humidity. A porous silicon (PS) relative humidity sensor has been used to illustrate the compensation procedure in order to evaluate the performance of the proposed method. The simulation results clearly show that the SVM technique can be effectively used to minimize the sensor errors due to hysteresis and nonlinearity. Compared with the results obtained from the back-propagation neural network (BPNN) technique, the proposed compensation method exhibited the best performance in performing compensation of a relative humidity sensor due to its remarkable generalization properties. It is observed from simulation studies that the hysteresis error may be eliminated. A desired linear relationship between the actual humidity and the SVM output can be obtained with proper SVM compensation. The root mean squared errors of the training-set or the testing-set are about 0.01. Moreover, based on periodic training of the flexible SVM, the proposed method has advantages in compensating the sensor long-term drift due to the aging effect and environmental changes. This method can also be extended to other sensors with hysteresis and nonlinearity errors.