Predicting Lithium-Ion Battery Health Status Using LSTM-Based Models

Accurate prediction of the state of health (SOH) of lithium-ion batteries is essential for ensuring the safety and reliability of electric vehicles. This study presents a practical approach using long short-term memory (LSTM) neural networks to estimate battery SOH based on real-world charge-discharge cycling data. We collected empirical measurements from 18650 lithium-ion cells under controlled cycling conditions, extracted key features, and applied several deep learning models, including two-layer LSTM, BiLSTM, CNN-LSTM, and CNN-LSTM with attention. Our results demonstrate that LSTM-based models can effectively capture SOH degradation trends, though their generalization is limited by data diversity and variations in battery chemistry. We discuss the implications of these findings and suggest future directions for improving model robustness and practical deployment in electric vehicle battery management systems.