Open AccessFast Remaining Capacity Estimation for Lithium‐ion Batteries Based on Short‐time Pulse Test and Gaussian Process Regression
Author(s) -
Ran Aihua,
Cheng Ming,
Chen Shuxiao,
Liang Zheng,
Zhou Zihao,
Zhou Guangmin,
Kang Feiyu,
Zhang Xuan,
Li Baohua,
Wei Guodan
Publication year - 2023
Publication title -
energy and environmental materials
Language(s) - English
Resource type - Journals
ISSN - 2575-0356
DOI - 10.1002/eem2.12386
Subject(s) - state of charge , kriging , battery (electricity) , gaussian , gaussian process , voltage , process (computing) , computer science , test data , simulation , engineering , machine learning , electrical engineering , power (physics) , chemistry , physics , quantum mechanics , programming language , operating system , computational chemistry
It remains challenging to effectively estimate the remaining capacity of the secondary lithium‐ion batteries that have been widely adopted for consumer electronics, energy storage, and electric vehicles. Herein, by integrating regular real‐time current short pulse tests with data‐driven Gaussian process regression algorithm, an efficient battery estimation has been successfully developed and validated for batteries with capacity ranging from 100% of the state of health (SOH) to below 50%, reaching an average accuracy as high as 95%. Interestingly, the proposed pulse test strategy for battery capacity measurement could reduce test time by more than 80% compared with regular long charge/discharge tests. The short‐term features of the current pulse test were selected for an optimal training process. Data at different voltage stages and state of charge (SOC) are collected and explored to find the most suitable estimation model. In particular, we explore the validity of five different machine‐learning methods for estimating capacity driven by pulse features, whereas Gaussian process regression with Matern kernel performs the best, providing guidance for future exploration. The new strategy of combining short pulse tests with machine‐learning algorithms could further open window for efficiently forecasting lithium‐ion battery remaining capacity.