PEMFC Output Voltage Prediction Based on Different Machine Learning Regression Models

Various machine learning models have been widely used in proton exchange membrane fuel cell performance prediction, life diagnosis, and other aspects. However, few studies have compared and analyzed the prediction effects of different models. In this work, four different classical machine learning models (Linear regression/Gaussian process regression/Support vector regression/Artificial neutral network) are trained to predict output voltage under specific operating conditions. The prediction effects of 4 models are compared and analyzed. The results show that the prediction effects of the four models are as follows from high to low: Gaussian process regression>Support vector regression>Artificial neutral network>Linear regression. Due to lack of sample at low current density in the training dataset, all 4 machine learning models own large prediction error near low current density zone. For predicting polarization curves, the Gaussian process regression and Artificial neutral network model shows better performance than the other two models. And especially, the nonlinear character could be expressed by the Gaussian process regression model.