A method for state of health estimation of lithium-ion batteries based on heterogeneous learner fusion and modal decomposition reconstruction

Accurate and reliable state-of-health (SOH) prediction is pivotal for the effective operation of battery management systems. However, conventional stacking models that directly pass base learner outputs to a meta-learner are prone to data redundancy and overfitting, ultimately compromising predictiv...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of power sources Jg. 655; S. 237932
Hauptverfasser: Tang, Hongyan, Chen, Yan, Zhou, Chengyu, Qi, Ziyi, Guo, Shuang
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.11.2025
Schlagworte:
Improved zebra optimization algorithm
Lithium-ion battery
Stacked ensemble learning
State of health
Variational mode decomposition
Lithium-ion battery
Improved zebra optimization algorithm
Variational mode decomposition
Stacked ensemble learning
State of health
ISSN:0378-7753
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Accurate and reliable state-of-health (SOH) prediction is pivotal for the effective operation of battery management systems. However, conventional stacking models that directly pass base learner outputs to a meta-learner are prone to data redundancy and overfitting, ultimately compromising predictive performance in complex scenarios. To address these limitations, This study proposes an SOH estimation method based on heterogeneous learning fusion with modal decomposition and reconstruction. Specifically, variational mode decomposition (VMD) is employed to denoise and reconstruct the preliminary outputs of base learners, thereby enhancing the representational quality of the input data. A deep extreme learning machine (DELM) is subsequently used to fuse the reconstructed signals with original input features for final SOH estimation. To overcome the inherent randomness in DELM initialization, an improved zebra optimization algorithm (IZOA)is introduced to optimize its parameters efficiently. The experimental results show that, on the CALCE dataset, the proposed method achieves SOH estimation errors within 1 %. Additionally, under small-sample conditions, the RMSE is below 0.623 % and the MAE is below 0.544 %, underscoring its high predictive accuracy and robust generalization capability. •VMD decomposes and reconstructs the prediction results of the optimized base learner.•A multi-strategy improved zebra optimization algorithm is proposed.•The proposed method is validated on NASA and CALCE datasets.•Introduce SHAP to improve model interpretability.
ISSN:0378-7753
DOI:10.1016/j.jpowsour.2025.237932