Electrical and Thermal Time Constants Fuel Cell System Identification: A Linear Versus Neural Network Approach

Abstract

The energy generated by PEM fuel cells can be used in many different applications with emphasis to commercial power generation and automotive application. It requires the integration of various subsystems such as chemical, mechanical, fluid, thermal and electrical ones. Their electrical and thermal time constants are important variables to analyze and consider in the development of control strategies of electronic converters. For this purpose, a mathematical model of the PEM fuel cell system was developed in Matlab/Simulink based on a set of equations describing cell operation. The model considers static and dynamic operating conditions of the PEM. Using experimental measurements at different load conditions made in a Nexa™ PEM fuel cell system, analysis based on linear ARX (Autoregressive with Exogenous Input) and neural network methods were made in Matlab in order to identify the electrical and thermal time constant values. Both linear ARX and neural network approaches can successfully predict the values of the time constants variables. However, the identification by the linear ARX is appropriated around the most significant operation points of the PEM system while neural network allows at obtaining a nonlinear global model. The paper intends to be a contribution for the identification of the electrical and thermal time constants of PEM fuel cells through these two methodologies. The linear approach is simple but presents some limitations while the non-linear one is widespread but more complex to be implemented.