Neural network modeling strategy applied to a multi-stack PEM fuel cell system

This work proposes applying a modeling methodology based on recurrent neural networks to a multi-stack fuel cell system composed of four Proton Exchange Membrane Fuel Cell (PEMFC) stacks. Even if the stacks have the same rated power and are from the same manufacturer, very often they present different performances (voltage response, efficiency and power curves). In this way, a model able to predict the behavior of each stack is necessary to guarantee an optimized operation of the whole system. Hence, the aforementioned methodology is used to obtain a prediction model for each stack aiming at their final application in a predictive control system. The models are also able to predict the power availability of the multi-stack system, being useful to be employed in the prognostics of the performance of the system in a vehicular application.