Coupling deep learning and multi-objective genetic algorithms to achieve high performance and durability of direct internal reforming solid oxide fuel cell

Abstract

Direct internal reforming (DIR) operation of solid oxide fuel cell (SOFC) reduces system complexity, improves system efficiency but increases the risk of carbon deposition which can reduce the system performance and durability. In this study, a novel framework that combines a multi-physics model, deep learning, and multi-objective optimization algorithms is proposed for improving SOFC performance and minimizing carbon deposition. The sensitive operating parameters are identified by performing a global sensitivity analysis. The results of parameter analysis highlight the effects of overall temperature distribution and methane flux on carbon deposition. It is also found that the reduction of carbon deposition is accompanied by a decrease in cell performance. Besides, it is found that the coupling effects of electrochemical and chemical reactions cause a higher temperature gradient. Based on the parametric simulations, multi-objective optimization is conducted by applying a deep learning-based surrogate model as the fitness function. The optimization results are presented by the Pareto fronts under different temperature gradient constraints. The Pareto optimal solution set of operating points allows a significant reduction in carbon deposition while maintaining a high power density and a safe maximum temperature gradient, increasing cell durability. This novel approach is demonstrated to be powerful for the optimization of SOFC and other energy conversion devices.