An online adaptive model for the nonlinear dynamics of fuel cell voltage

Polymer electrolyte fuel cells have been widely used in automotive applications, in which fast-response and highly accurate fuel cell systems are required to achieve good performance. To fulfill this requirement, an adaptive fuel cell model is developed herein for a polymer electrolyte fuel cell system. The model is established on the basis of a least squares support vector machine. A genetic algorithm is employed to set the initial values of the internal parameters of the model by incorporating existing data from previous experiments. Then, an adaptive process is further conducted to provide an online update of the model’s internal parameters. The genetic algorithm can effectively avoid the initial parameters by falling to a local minimum. Moreover, the online updating of the parameters makes the model more adaptive to load changes in the real-time application of the fuel cell system. The proposed model is experimentally-tested on a fuel cell test rig. The results indicate that the proposed model can accurately and effectively predict fuel cell voltage. In addition, two reference models are employed to compare with the online adaptive model, by which the advantages of the genetic algorithm and parameter updating are verified. The model accuracy is improved significantly with the genetic algorithm, indicating the importance of initial parameters setting. The gradient method also benefits the model’s accuracy in online modeling and predicting, but its efficiency still depends on the initial parameters. This online adaptive model can easily address frequent load change and the long term operation of fuel cells.