Modeling of pem fuel cells by using artificial neural networks

Fuel cell parameters are difficult to determine for fuel cell systems because of their non-linearity. The modeling of the proton exchange membrane (PEM) fuel cell requires amultidisciplinary approach including the study of electrochemistry, polymer science, heattransfer, fluid dynamics and mass transfer. In the literature, modeling of PEM fuel cells areusually done with complex models based on a good knowledge of physicochemicalphenomena.Artificial Neural Network (ANN) is an information processing paradigm that is inspired bybiological nervous systems. The ability of ANN to represent non-linear systems makes it apowerful tool for modeling. The purpose of this thesis is to review literature on PEM fuelcell modeling, and derive a non-parametric empirical model including process variations toestimate the performance of polymer electrolyte membrane fuel cells by using artificialneural networks.In this study, firstly, information about artificial neural networks and PEM fuel cells aregiven. Then, PEM fuel cell modeling techniques are presented, and an artificial fuel cellmodel proposed. The model uses a ANN which has eight input and an output. To producethe correct output data, the network was trained with improved versions of the BackPropagation algorithm, the Levenberg-Marquardt and Quasi-Newton algorithm. Half of theoperational points were used to train the ANN Model, while the other half was used for thevalidation. During the learning process, the error function was minimised with an increasingnumber of training epochs. After the final training, the ANN was ready to generaterelationship between inputs and outputs. The average values of the absolute errors is wellbelow 1 %, and the maximum error is arround 4%.Key Words: PEM fuel cells, artificial neural networks, modeling, Levenberg-Marquardtlearning algorithm, Quasi-Newton learning algorithm