Optimization Framework for Redox Flow Battery Electrodes with Improved Microstructural Characteristics

Abstract

This research significantly advances the field of vanadium redox flow batteries (VRFBs) by introducing a pioneering approach to optimize the microstructural characteristics of carbon electrodes. Addressing the traditional challenge of developing high-performance electrode materials for VRFBs, this study employs a robust, generalizable, and cost-effective data-driven modeling and optimization framework. A novel sampling strategy using low-discrepancy Latin Hypercube and quasi-Monte Carlo methods generates a small-scale, high-fidelity dataset with essential space-filling qualities for training supervised machine learning models. This study goes beyond conventional methods by constructing two surrogate models: a random forest regressor and a gradient boosting regressor as objective functions for optimization. The integration of a non-dominated sorting genetic algorithm II (NSGA-II) for multi-objective optimization facilitates exhaustive exploration of the surrogate models, leading to the identification of electrode designs that yield enhanced energy efficiencies (EEs) under specific operating conditions. The application of NSGA-II in exploring surrogate models not only facilitates the discovery of realistic design combinations but also adeptly manages trade-offs between features. Based on the insights obtained from the optimization framework, new electrode types were fabricated, demonstrating marked improvements in EE and validating the machine-learning recommendations. This research highlights the critical role of improved electrical conductivity and porosity in carbon materials, showing their direct correlation with increased EE. Overall, this study represents a significant step forward in developing more efficient and practical VRFBs, offering a valuable contribution to the renewable energy storage landscape.