Dynamic modelling of a polymer electrolyte membrane fuel cell stack by nonlinear system identification

Real-time models of polymer electrolyte membrane fuel cell (PEMFC) stacks with high accuracy are required, e.g. for the design of controllers or online diagnosis tools. By using physical and chemical laws representing the processes in a PEMFC stack, very detailed, but computationally complex models can be retrieved. In this paper, a nonlinear dynamic model obtained by system identification is proposed for PEMFC stacks. The model structure in this contribution is based on a modular concept and is divided into a static and a dynamic part. The static part represents the stationary points and the dynamic part describes the deviation from these stationary points due to changes in the input signals. Both parts can be modelled by different methods. A characteristic map and a neural network (NN) are proposed for the static part. For the dynamic part, transfer functions and a linear state-space model retrieved by canonical variate analysis (CVA) are investigated.