Performance prediction and power density maximization of a proton exchange membrane fuel cell based on deep belief network

Abstract

Intelligent methods have become powerful modeling tools for predicting the performance of complex systems. As one of these methods, the deep belief network (DBN) is employed to construct the data-driven model for predicting proton exchange membrane fuel cell (PEMFC) performance. First, a three-dimensional PEMFC numerical model is developed as the source of DBN training data. Second, the DBN model is established to fit the simulation data, and the DBN hyper-parameters are determined by cross-validation. After the DBN model is validated, it is compared with back-propagation (BP) neural network and Bagging neural network ensemble model to check its prediction performance. Finally, the DBN model is applied to analyze effects of network input variables on PEMFC performance, and the optimal operating condition is obtained. Results show that the DBN model can predict the PEMFC current density precisely, and the DBN prediction accuracy is superior to that of other intelligent methods. The DBN model can acquire the optimal operating condition corresponding to the maximum power density. Therefore, the DBN method is an effective tool to model PEMFC systems.