A long sequence synthetic battery parameter generation perspective using reliable self‐attention mechanism.

Summary: The automotive sector around the world is undergoing a massive transition towards using cleaner and sustainable forms of energy. Lithium‐ion batteries are the key driver for this ongoing electrification transformation of all modes of transportation. Range computation of all such electric vehicles hinges on precise State‐of‐Charge (SOC) estimation of battery packs. Although significant research endeavours have focused on developing a series of SOC estimation techniques, accessibility to high‐quality battery parameter data (Voltage, Current and Temperature) still remains a challenge. Moreover, there are very few diverse open access datasets available at the moment. This paper directs its efforts in generating manifold battery datasets which result in improved performance of Deep learning architectures. The key contributions of this work are 2fold: (a) we highlight the effectiveness of the pyramidal self‐attention approach for producing reliable synthetic data, (b) we illustrate the generalization capability of this technique by evaluating this approach over different battery chemistries. The attention mechanism resulted in a mean absolute error of 0.3 and 0.2 for the pouch and cylindrical cell respectively. We provide implementation details in our results section which will empower researchers to reproduce our work effectively. [ABSTRACT FROM AUTHOR]