Multi-objective optimization of solid oxide electrolysis cell efficiency and thermal gradient using an active-learning–CFD hybrid framework

High-fidelity three-dimensional computational fluid dynamics (CFD) provides detailed insight into solid oxide electrolysis cell (SOEC) behavior. However, the high computational cost associated with such simulations significantly limits the efficient exploration and optimization of operating conditions. In this study, a closed-loop optimization framework combining CFD, a Gaussian process (GP) surrogate model, and epsilon-Pareto Active Learning (ε-PAL) was developed for efficient SOEC operation optimization. Electrochemical performance index (EPI) and in-plane temperature difference were selected as the key objectives representing electrochemical performance and thermal stability. Compared to the random sampling method, the ε-PAL strategy achieves a 2.5% higher final EPI and a 90.5% lower final temperature difference. Moreover, compared with a full 6561-point grid search requiring 22,963.5 h of computation (3.5 h per case), the ε-PAL strategy achieved comparable optimization performance using only 17 CFD evaluations, corresponding to a total computational cost of just 60 h.