Treffer: High-Fidelity SOH Prediction in Lithium-Ion Batteries Using Hybrid ML Networks

Accurate and efficient prediction of lithium-ion battery state of health (SOH) is critical for ensuring reliability in electric vehicles, grid storage, and aerospace systems. Traditional SOH estimation methods often struggle with nonlinear degradation behaviors and lack sensitivity to subtle electrochemical signals, limiting their real-world deployment. To address these challenges, this study examines hybrid deep learning models that integrate differential capacity (dQ/dV) analysis to enhance predictive accuracy. Four hybrid architectures - hybrid CNN-LSTM multihead, CNN extractor for LSTM, DNN-LSTM, and DNN Bi-LSTM - were developed and evaluated using the NASA randomized battery usage dataset, offering a realistic benchmark under diverse operational profiles. Among these, the hybrid DNN-LSTM model achieved the best performance, with high predictive accuracy (R² = 0.9968, MAE = 0.63%) and computational efficiency, making it well-suited for real-time battery management applications. Its lightweight design allows rapid adaptation to different chemistries and usage conditions, with potential for remaining useful life (RUL) estimation and diagnostics. This study highlights the advantages of combining dQ/dV analysis with hybrid deep learning architectures, providing a scalable and practical solution for modern battery health management.