Researchers have proposed an efficient feature search approach for estimating the state of health of lithium-ion batteries, aiming to reduce the reliance on manually selected aging features. The method uses Bayesian optimization to search multidimensional feature spaces and then applies an ensemble regression model to improve the accuracy and robustness of battery health estimation.
Accurate state of health, or SOH, estimation is fundamental for battery prognostics and health management. In electric vehicles, energy storage systems, and other battery-powered technologies, operators need to know not only how much charge a battery currently holds, but also how its usable capacity is changing over time. Poor SOH estimation can affect maintenance planning, safety assessment, second-life screening, and the design of battery management strategies.
A major challenge is that many existing SOH estimation methods depend heavily on manually selected features. Researchers or engineers often choose aging indicators based on prior experience, domain knowledge, or trial-and-error analysis. While this can work well in specific settings, it may also introduce subjective bias and limit generalizability. A feature that performs well for one dataset, chemistry, or operating condition may not remain reliable across broader battery populations.
The new study addresses this limitation by proposing a feature search method for lithium-ion batteries in multidimensional feature spaces, including Cycle-Voltage-SOC. According to the article, the method evaluates features based on their self-volatility and correlation with battery capacity. This design is intended to identify features that are both informative and efficient, rather than relying only on human judgment to decide which parts of the battery data should be used.
A key component of the framework is an evaluation index designed to characterize feature efficiency. This allows the search process to consider whether a candidate feature is useful for SOH estimation while also accounting for its behavior across battery aging data. By making feature quality part of a structured search process, the approach aims to improve interpretability and reduce the uncertainty introduced by manual feature engineering.
The researchers then apply Bayesian optimization to search for optimal features efficiently. Bayesian optimization is useful in this context because feature spaces can be large, and testing every possible feature combination may be impractical. Instead of exhaustively scanning the entire space, the method guides the search toward promising regions. This can reduce the dependence on human experience and make feature extraction more systematic.
With the searched features, the study develops a stacking ensemble regression model to estimate battery SOH. The paper highlights an integrated model based on Bayesian ridge regression and random sampling consensus for SOH estimation. Such an ensemble approach can help improve robustness by combining complementary modeling strengths and reducing the effect of noisy or unstable data. In practical battery applications, this is important because aging measurements may vary across cycles, cells, and operating conditions.
Experimental results reported in the paper demonstrate the efficiency of both the proposed feature search method and the SOH estimation model. The study states that the approach provides interpretable feature extraction and improves SOH estimation accuracy. While the ScienceDirect summary does not provide detailed numerical error values, the reported results indicate that automated feature search can strengthen battery health estimation by selecting more reliable aging indicators from multidimensional data.
Further work will still be needed to evaluate the method across wider battery chemistries, application scenarios, temperature conditions, and real-world operating profiles. Even so, the study offers a strong indication that combining feature-efficiency evaluation, Bayesian optimization, and robust ensemble regression could make lithium-ion battery SOH estimation less dependent on subjective feature selection. For battery management and prognostics, that could support more reliable decisions about maintenance, reuse, and long-term energy-system operation.
Reference
Author:
Xiaolong Chen a , Huihui Yang a , Yingze Yang a , Heng Li a , Yuan Chen b , Chaolong Zhang c , Longxing Wu d , Lisen Yan e
Title of original paper:
An efficient feature search approach for robust state of health estimation of Li-ion battery
Article link:
https://www.sciencedirect.com/science/article/pii/S2773153725000891
Journal:
Green Energy and Intelligent Transportation
DOI:
10.1016/j.geits.2025.100339
Affiliations:
a School of Electronic Information, Central South University, Changsha 410004, China
b School of Artificial Intelligence, Anhui University, Hefei 230601, China
c College of Intelligent Science and Control Engineering, Jinling Institute of Technology, Nanjing 211199, China
d College of Intelligent Manufacturing, Anhui Science and Technology University, Chuzhou 233100, China
e Center for Ageing, Reliability and Lifetime Prediction of Electrochemical and Power Electronics Systems (CARL), RWTH Aachen University, Campus-Boulevard 89, Aachen 52074, Germany
Green Energy and Intelligent Transportation
Experimental study
Not applicable
An efficient feature search approach for robust state of health estimation of Li-ion battery
5-Feb-2026