Integration of supervised machine learning for predictive evaluation of chemical looping hydrogen production and storage system

Abstract

Chemical looping technology is an emerging method for hydrogen production and storage, characterized by its environmentally friendly and safe inherent gas separation processes. However, the development of this technology requires consideration of oxygen carrier selection, reactor design, and process optimization, trial-and-error experimental methods are labor-intensive and costly. Herein, we propose the integration of machine learning into the chemical looping hydrogen production system to achieve accurate prediction and evaluation during the development process. Based on a dataset of 315 data sets, the ANN and Extra Tree models demonstrated the highest generalization ability among six models, with prediction accuracies for hydrogen yield and purity reaching R² = 0.96 and R² = 0.94, respectively. The interpretability algorithm analyzed the impact of different input parameters on hydrogen yield and purity, revealing that reaction temperature and fuel gas had the most significant influence. We predicted the hydrogen production performance of four new-input natural oxygen carriers using the trained ANN and Extra Tree models. The results indicated that the predictions were generally consistent with experimental results, with the best oxygen carrier maintaining a hydrogen yield of ∼3.12 mmol g⁻¹ and a hydrogen purity of 99.65% after 10 cycles. In summary, machine learning can serve as an alternative to traditional trial-and-error methods, accelerating the development process of chemical looping hydrogen production technology.