Evaluation of the parameters of a PEM fuel cell system by using machine learning regression models

Abstract

Proton exchange membrane fuel cells (PEMFCs) operate with various parametric values, among which temperature, pressure, and flow rate are the most prominent. The effects of these parameters on the current density of PEMFCs and their interactions with each other are important. This study performed a comprehensive parametric evaluation. For this purpose, PEMFC stack simulation was conducted, and the effects of parameters on current density were determined. Various machine learning (ML) methods were applied to the obtained results, and regression analysis was performed. Current density was considered the output, and operating temperature, fuel flow rate, air flow rate, and fuel supply pressure were regarded as inputs. In total, 24 ML methods were applied, and 6 of them yielded excellent results. The six ML methods were the fine tree model, multilinear regression, cubic support vector machine, the bagged tree model, bilayered neural network, and rational quadratic Gaussian process regression (GPR). The regression coefficient (R2), mean absolute error, mean square error, and root mean square error of each model were derived. Results indicated that rational quadratic GPR and bilayered neural network were the most effective among all the methods. The R2 values of the rational quadratic GPR and bilayered neural network models were equal to 1, indicating a perfect match between the predicted and observed values.