Optimization of maximum power density output for proton exchange membrane fuel cell based on a data-driven surrogate model

Abstract

Operating conditions are of great significance for proton exchange membrane fuel cells (PEMFCs) and directly determine the output performance of PEMFC. In this paper, an optimization framework is proposed, which is combining a neural network data-driven surrogate model and a stochastic optimization algorithm to achieve multi-variable global optimization, and optimizes the operating conditions for enhancing the power density of the PEMFC. A numerical model of the PEMFC is constructed as the source of the database, and the database is used to train a data-driven surrogate model based on radial basis functions (RBF), which is a typical feed-forward neural network. Then the surrogate model is fed into a particle swarm optimization (PSO) algorithm to obtain the optimal solution for the best combination of operating conditions. Results show that the surrogate model can accurately predict the output voltage of the PEMFC model, where the squared correction factor (R-square) and the mean percentage error of the test set are 0.99638 and 4.4686% respectively. And the optimization framework using a combination of data-driven surrogate model and stochastic optimization algorithm can obtain the maximum power density of 0.6097 W cm−2, with a relative error of only 1.5321% to the PEMFC model results. The result shows that the framework can effectively handle the multivariate optimization of complex systems.