Machine learning-genetic algorithm coupled optimization of heteroatom-doped carbon electrodes for vanadium redox flow batteries

The development and optimization of high-performance carbon electrodes are critical for vanadium redox flow batteries (VRFBs). However, electrode performance is influenced by multiple interacting factors, while traditional experimental approaches are often time-consuming, inefficient, and costly. This study establishes a data-driven framework that integrates machine learning with genetic algorithms, enabling accurate prediction and global optimization of the performance of heteroatom-doped carbon electrodes under diverse conditions. A multidimensional database is constructed based on published literature and active learning, followed by the training and evaluation of multiple machine learning models. The optimal model achieves excellent predictive accuracy, with R 2 values of 0.982 and 0.991 for coulombic efficiency and voltage efficiency, respectively. Shapley additive explanations (SHAP) analysis further reveals the underlying regulatory mechanisms of key parameters affecting electrode performance. Moreover, an elite genetic algorithm is employed to screen optimal electrode performance. The optimized electrode exhibits a predicted energy efficiency of 85.2% at 200 mA cm −2 with outstanding cycling stability. Experimental results show strong agreement with model predictions, with the mean absolute percentage error remaining below 3.7%. This work provides an efficient and reliable strategy for the rational design of doped electrodes, which significantly accelerates the development of next-generation redox flow batteries.