A Neural Non-linear Predictive Control for PEM-FC

This paper deals with the application of non-linear predictive control with neural networks to Proton Exchange Membrane Fuel Cells (PEM-FC). The control objective is to regulate the cell voltage, acting on the hydrogen pressure, trying to reduce the variation of the input control variable. An analysis of the non-linearities of the fuel cell stack has been carried out, making use of a suitable fuel cell model. The non-linear predictive control has been implemented by several neural networks (multi value perceptrons), after dividing the operating domain into three areas according to the cell current value (low loads, quasi-linear zone and high loads). Simulation results have been provided and discussed, showing the goodness of the proposed non-linear control technique in reducing the variations of hydrogen pressure.