MambaLithium: Selective state space model for remaining-useful-life, state-of-health, and state-of-charge estimation of lithium-ion batteries
Zhuangwei Shi
TL;DR
Leveraging Mamba algorithms, MambaLithium adeptly captures the intricate aging and charging dynamics of lithium-ion batteries, and not only enhances estimation accuracy but also maintains computational robustness.
Abstract
Recently, lithium-ion batteries occupy a pivotal position in the realm of electric vehicles and the burgeoning new energy industry. Their performance is heavily dependent on three core states: remaining-useful-life (RUL), state-of-health (SOH), and state-of-charge (SOC). Given the remarkable success of Mamba (Structured state space sequence models with selection mechanism and scan module, S6) in sequence modeling tasks, this paper introduces MambaLithium, a selective state space model tailored for precise estimation of these critical battery states. Leveraging Mamba algorithms, MambaLithium adeptly captures the intricate aging and charging dynamics of lithium-ion batteries. By focusing on pivotal states within the battery's operational envelope, MambaLithium not only enhances estimation accuracy but also maintains computational robustness. Experiments conducted using real-world battery data have validated the model's superiority in predicting battery health and performance metrics, surpassing current methods. The proposed MambaLithium framework is potential for applications in advancing battery management systems and fostering sustainable energy storage solutions. Source code is available at https://github.com/zshicode/MambaLithium.