Active learning based reverse design of hydrogen production from biomass fuel

Abstract

Biomass hydrogen production holds substantial promise in addressing critical challenges within the realms of renewable energy and environmental sustainability. Nevertheless, the labor- and time-intensive nature of conducting combinatorial experiments and relying on traditional trial-and-error methods significantly constrains the exploration of biomass fuel solutions. This paper introduces a novel framework for the reverse engineering of hydrogen production systems, which seamlessly integrates Bayesian active learning with machine learning models. In Bayesian active learning, 307 biomass fuel samples were expanded to 1354 in order to ensure the accuracy of the model prediction between the seven characteristics of biomass fuel and hydrogen production. Four machine learning algorithms including RF, XGBoost, KNNR, AdaBoost were used to build the prediction model. A stochastic search reverse design approach based on the XGBoost model with the highest coefficient of determination (R2 = 0.91) was innovatively used to realize the “reverse design” from “target performance” to “composition and process”. Through reverse design, a two-order-of-magnitude increase in design space is achieved in hundreds of times less time than conventional design time, resulting in a significant improvement in hydrogen production efficiency. The practical application of this method for hydrogen production from biomass fuels may go far beyond linear extrapolation based on the current state of the art and can be applied to a wider range of material designs.