Fig. 3: ARCANA’s predictive performance on cylindrical sample cells.
From: Attention towards chemistry agnostic and explainable battery lifetime prediction
The performance of the proposed framework on two unseen datasets, namely cylindrical DNMC+NCA in Panel I and prismatic DLCO in Panel II, when predicting battery behavior over 500 cycles for three predictors of Voltage drop [V] (a), CE (b) and Qdis [Ah] (c). The uncertainty at the 10th and 90th percentiles effectively captures underlying data variability and highlights the model’s predictive reliability and adaptability across diverse unseen datasets, demonstrating deep insight into data characteristics.
