This work deals with the modeling of a PEM fuel cell generator. This item will be seen here under the angle of Artificial Neural Network and applied to two different PEM fuel cells. By training, Artificial Neural Networks enable to carry out models which adapt to experimental behaviours observed. The first part talks about the technological problems bonded to the fuel cell integration in a vehicle. Then, the author points out the necessity of modelling a fuel cell generator before listing the different way of modelling. The first step of this study is dedicated to the possibility to realize a Neural Network model able to evaluate the static mode of a PEM fuel cell. The second part describes the approach which allows carrying out this first model. This part is divided in three essential points; choice of a network structure, choice of experimental tests to establish a representative training sequence of the system and choice of inputs/outputs of the model, study of different training methods carrying out to a good modeling. The conclusion of this part emphasizes the extrapolation ability of this model to more powerful fuel cell. To have a complete model, dynamic behaviour of the fuel cell must be studied. The development of the dynamic model using recurrent neural network is presented in a third part. In order to conclude this work, an original method based on Fourier analysis is proposed. By this way, in order to predict the voltage response of the fuel cell under dynamic current solicitations, a multi model black box, which couples dynamic and static models, has been developed. Finally, a sensitive parametric analysis is done.
