Modeling of commercial proton exchange membrane fuel cell using support vector machine

Azadeh Kheirandish, Niusha Shafiabady, Mahidzal Dahari, Mohammad Saeed Kazemi, Dino Isa

Research output: Contribution to journalArticlepeer-review


A method for predicting the performance of a proton exchange membrane fuel cell (PEMFC) system of a commercially available electrical bicycle using a support vector machine (SVM) is presented in this paper. The main advantage of the results obtained from this study is facilitating the use of carbon-free fuels instead of carbon-based ones and consequently reducing the energy consumption. Because such cells are nonlinear, multivariable systems that are difficult to model through traditional methods hence SVMs, which are powerful tools for predicting PEMFC performance, are used. Experimental data obtained from a 250 W PEMFC were used to predict parameters to describe the V-I, P-I, and efficiency-power curves, and then, the data was applied to predict overall PEMFC performance. To evaluate the functionality of suggested model, this method has been compared with multi-layer perceptron (MLP) artificial neural network model. It has been demonstrated that, the error of SVM model is much smaller than MLP, and the proposed approach has better performance to predict the PEM fuel cell curve for the electrical bicycle. It was shown that the coefficient of determination in the SVM prediction model for power-current curve is approximately 99%, which is 97% for MLP model that makes the proposed black box SVM PEMFC model suitable for monitoring and simulating fuel cell performance in the electrical bicycle that is beneficial for its variety of energy saving applications.

Original languageEnglish
Pages (from-to)11351-11358
Number of pages8
JournalInternational Journal of Hydrogen Energy
Issue number26
Publication statusPublished - 13 Jul 2016
Externally publishedYes


Dive into the research topics of 'Modeling of commercial proton exchange membrane fuel cell using support vector machine'. Together they form a unique fingerprint.

Cite this