Comparative Study on the Performance Prediction of Fuel Cell Using Support Vector Machine with Different Kernel Functions

Abstract

The performance attenuation characteristics of fuel cell stack show a strong nonlinearity, but there is still no good model to predict its nonlinear attenuation characteristics. Support vector regression (SVR) is used as linear regression algorithm and realizes nonlinear regression function by introducing kernel function. Based on the test data of 4 kW fuel cell stack running 600 h on the bench, the results of using the SVR model with Gaussian radial basis kernel function (G-RBF), sigmoid kernel function, polynomial kernel function, and mixed kernel function to predict fuel cell performance are compared. The results show that the SVR model with polynomial kernel function has higher prediction precision.