Analysis of the time-varying behavior of a PEM fuel cell stack and dynamical modeling by recurrent neural networks

This work presents an analysis of the time-varying behavior of a PEM fuel cell (PEMFC) stack based on experimental results, pointing out some constraints that should be taken into account in the development of a control system for the stack. A system identification methodology based on recurrent neural networks is proposed to model such behavior. A dynamic model using this technique is developed for a commercial PEMFC stack operating under a real load profile. The results show that the neural model is able to track the stack voltage dynamics with a very low error.