Few-shot learning of initial coulombic efficiency using adaptive combination kernel deep Gaussian process regression

Hai Guo, Hongcheng Zhang, Xiaofeng Lv, Xiaoxu Liu, Tianyi Ji,
Few-shot learning of initial coulombic efficiency using adaptive combination kernel deep Gaussian process regression,
Journal of Energy Storage,
Volume 128,
2025,
117238,
ISSN 2352-152X,
https://doi.org/10.1016/j.est.2025.117238.
(https://www.sciencedirect.com/science/article/pii/S2352152X25019516)
Abstract: The initial coulombic efficiency (ICE) is critical for minimizing active ion loss and enhancing battery stability. However, the complex interplay between the multiscale structure of electrode materials and manufacturing processes obscures the underlying optimization mechanisms, hindering the development of high-performance batteries. Herein, a machine learning method based on adaptive combinatorial kernel deep Gaussian process (AKC-DGPR) is proposed on 110 sets of experimentally collected datasets, including the preparation process, structural parameters, and ICEs of hard carbon. A few-shot learning framework is developed especially for laboratory environments where data are collected manually. The kernel functions are first collected and organized to construct the kernel function list. The global optimal solution is gradually approached through iteration, and the SHapley Additive exPlanations (SHAP) method is used for model interpretation. The results show that the R2 of the model is 0.74, and the RMSE is 0.13. The proposed model has high accuracy and reliability in predicting ICE, and the SHAP method explains the effect of input features on ICE and makes the model's decision-making process transparent. An inverse design idea is further proposed based on this model, which provides solid support for few-shot data-driven SIB material design and performance optimization.
Keywords: Initial coulombic efficiency; Sodium-ion battery; Deep Gaussian process; Few-shot learning; SHAP analysis