Challenge
State of Charge and State of Health estimates for EV batteries must remain reliable across different charging and temperature conditions.
Home/Hybrid LSTM-XGBoost Stacking Ensemble for Multi-Condition Battery Management System Optimization in Electric Vehicles
2026 · AI/Research
Hybrid LSTM-XGBoost Stacking Ensemble for Multi-Condition Battery Management System Optimization in Electric Vehicles
A hybrid LSTM (128-64-32 unit feature extractor) and XGBoost (500 trees) ensemble that estimates State of Charge and State of Health of 18650 lithium-ion batteries in EVs, including 2C/3C fast charging and -10°C to 45°C temperatures.
Authors: I Made Prabu Mahendra Putra Rebawa, Michael Jemmy Tanzel, Nikita Ananda Putri Masaling, Edy Irwansyah (BINUS University)
Role: Researcher & AI Model Developer
Team: BINUS University
Stack: Python, LSTM, XGBoost, Stacking Ensemble
Solution
Used a 128-64-32-unit LSTM feature extractor followed by 500-tree XGBoost in a stacking ensemble to estimate SoC and SoH.
0,202%
SoC RMSE
ECM 7,945%
0,105%
SoH RMSE
1,000
Anomaly detection · F1-Score
Key features
Battery data analysis
Analyzes battery behavior across varied electric-vehicle conditions.
Stacking ensemble model
LSTM feature extraction combined with 500 XGBoost trees.
Outcome
The model maintained strong performance under fast-charging and extreme-temperature scenarios.