A recurrent neural approach for modeling non-reproducible behavior of PEM fuel cell stacks

Author

da Costa Lopes, F. ; Watanabe, E.H. ; Rolim, L.G.B.

Author_Institution

Dept. of Special Technol., CEPEL-Electr. Energy Res. Center, Rio de Janeiro, Brazil

fYear

2013

fDate

25-28 Feb. 2013

Firstpage

661

Lastpage

667

Abstract

This work presents a recurrent neural model for PEM fuel cell stacks based on NARX and NOE neural structures. The practical difficulties to employ electrochemical models and the causes of the odd behavior observed in some PEM stacks are discussed. The model developed predicts the terminal voltage of a stack even in the situation where it presents a non-reproducible behavior, such as unexpected voltage fluctuations. The predictive capacity of the model is evaluated in two different scenarios, showing good agreement with the measured data.

Keywords

electrochemical analysis; electrochemical electrodes; load forecasting; power engineering computing; proton exchange membrane fuel cells; recurrent neural nets; NARX neural structure; NOE neural structure; PEM fuel cell stack; electrochemical model; electrode; nonreproducible behavior modeling; predictive capacity model; recurrent neural model; terminal voltage prediction; unexpected voltage fluctuation; Current measurement; Fuel cells; Load modeling; Predictive models; Temperature measurement; Training; Voltage measurement; NARX neural network; NOE neural network; PEM fuel cell stack; modeling; voltage prediction;

fLanguage

English

Publisher

ieee

Conference_Titel

Industrial Technology (ICIT), 2013 IEEE International Conference on

Conference_Location

Cape Town

Print_ISBN

978-1-4673-4567-5

Electronic_ISBN

978-1-4673-4568-2

Type

conf

DOI

10.1109/ICIT.2013.6505750

Filename

6505750