Abstract
Intelligent methods have become powerful tools for modeling and optimization of complex systems. In this study, a Deep Belief Network (DBN) is employed to predict the current density of a Proton Exchange Membrane Fuel Cell (PEMFC), and a multi-objective optimization framework is proposed as well. Firstly, a single PEMFC CFD model is developed as the basic model, which is verified by the experiment data. Secondly, the DBN is employed to construct a performance prediction model of PEMFC. The DBN hyper-parameters are determined by cross validation method, and the DBN model is compared with other datadriven models to prove its superiority. Finally, a multi-objective optimization framework combining significant variables recognition, surrogate models and a multi-objective genetic algorithm is proposed. Results show that the DBN model predicts the PEMFC current density precisely, and the DBN prediction accuracy is superior to that of other intelligent methods. The multi-objective optimization framework can find the final Pareto front in only ten minutes, which improves the optimization efficiency significantly. This study provides efficient tools for PEMFC performance prediction and optimization, and can be a guide for engineering applications.
