A study into Proton Exchange Membrane Fuel Cell power and voltage prediction using Artificial Neural Network

Abstract

Polymer Electrolyte Membrane fuel cell (PEMFC) uses hydrogen as fuel to generate electricity and by-product water at relatively low operating temperatures, which is environmentally friendly. Since PEMFC performance characteristics are inherently nonlinear and related, predicting the best performance for the different operating conditions is essential to improve the system’s efficiency. Thus, modeling using artificial neural networks (ANN) to predict its performance can significantly improve the capabilities of handling multi-variable nonlinear performance of the PEMFC. This paper predicts the electrical performance of a PEMFC stack under various operating conditions. The four input terms for the 5 W PEMFC include anode and cathode pressures and flow rates. The model performances are based on ANN using two different learning algorithms to estimate the stack voltage and power. The models have shown consistently to be comparable to the experimental data. All models with at least five hidden neurons have coefficients of determination of 0.95 or higher. Meanwhile, the PEMFC voltage and power models have mean squared errors of less than 1 × 10−3 V and 1 × 10−3 W, respectively. Therefore, the model results demonstrate the potential use of ANN into the implementation of such models to predict the steady state behavior of the PEMFC system (not limited to polarization curves) for different operating conditions and help in the optimization process for achieving the best performance of the system.