AMiT: A Battery State-of-Health Prediction Method Integrating Physical Features and Optimization Algorithms

Lithium-ion batteries are widely used in energy storage and electrified transportation. Accurate state-of-health (SOH) prediction is essential for safe and reliable operation, yet variations in degradation trajectories and local capacity fluctuations complicate prediction from partial charging data. This study proposes AMiT, a multi-scale capacity prediction framework integrating physical-feature extraction, multi-scale decomposable mixing (MDM), an iTransformer backbone, and artificial lemming algorithm (ALA) hyperparameter optimization. Health indicators are extracted from local constant-current charging segments, and MDM fuses temporal representations at different resolutions. The iTransformer models dependencies among input variables, while hyperparameters are selected according to validation error. The framework is evaluated on eight batteries from the NASA, CALCE, Tongji, and XJTU public datasets and through four-fold leave-one-battery-out validation (LOBO) on CALCE. In within-cell chronological prediction over five random seeds, AMiT reduces reference-capacity-normalized root mean squared error (RMSE) by 31.26% and 57.67% relative to iTransformer on NASA B0005 and B0006, respectively. Across the four CALCE LOBO folds, the battery-equal macro-averaged reference-capacity-normalized RMSE is 1.5943 percentage points (pp), representing reductions of 4.86% and 6.94% relative to the temporal convolutional network (TCN) and iTransformer, respectively. In paired ablation experiments on NASA B0006, adding MDM reduces RMSE by 69.15%. On CALCE CS2-35, ALA reduces RMSE by 54.50% relative to random search with the same number of candidate evaluations. Future work will focus on reducing stage-dependent capacity bias and evaluating performance under varying temperatures and charging conditions.

Authors

Institutions

Publication Details

Journal
Energies
Published
2026-11-17
DOI
https://doi.org/10.3390/en19225561
Primary Topic
Battery Modeling and Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AMiT: A Battery State-of-Health Prediction Method Integrating Physical Features and Optimization Algorithms

Xiangchen Zhu, Yutong Liu, Jiangbin Tian, Bo Huang et al.
4 links
Energies
Battery Modeling and Management
article

AMiT: A Battery State-of-Health Prediction Method Integrating Physical Features and Optimization Algorithms

Xiangchen Zhu, Yutong Liu, Jiangbin Tian, Bo Huang, Guohui Zeng, Jinbin Zhao
article en

Abstract

Lithium-ion batteries are widely used in energy storage and electrified transportation. Accurate state-of-health (SOH) prediction is essential for safe and reliable operation, yet variations in degradation trajectories and local capacity fluctuations complicate prediction from partial charging data. This study proposes AMiT, a multi-scale capacity prediction framework integrating physical-feature extraction, multi-scale decomposable mixing (MDM), an iTransformer backbone, and artificial lemming algorithm (ALA) hyperparameter optimization. Health indicators are extracted from local constant-current charging segments, and MDM fuses temporal representations at different resolutions. The iTransformer models dependencies among input variables, while hyperparameters are selected according to validation error. The framework is evaluated on eight batteries from the NASA, CALCE, Tongji, and XJTU public datasets and through four-fold leave-one-battery-out validation (LOBO) on CALCE. In within-cell chronological prediction over five random seeds, AMiT reduces reference-capacity-normalized root mean squared error (RMSE) by 31.26% and 57.67% relative to iTransformer on NASA B0005 and B0006, respectively. Across the four CALCE LOBO folds, the battery-equal macro-averaged reference-capacity-normalized RMSE is 1.5943 percentage points (pp), representing reductions of 4.86% and 6.94% relative to the temporal convolutional network (TCN) and iTransformer, respectively. In paired ablation experiments on NASA B0006, adding MDM reduces RMSE by 69.15%. On CALCE CS2-35, ALA reduces RMSE by 54.50% relative to random search with the same number of candidate evaluations. Future work will focus on reducing stage-dependent capacity bias and evaluating performance under varying temperatures and charging conditions.

EnergiesVol. 19(22)
Shanghai University of Engineering Science (CN), Shanghai University of Electric Power (CN), Wind Power Engineering (Japan) (JP), Aalborg University (DK)
Affordable and clean energy, Industry, innovation and infrastructure
Battery Modeling and Management
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon