Electrochemically informed state-of-health estimation of lithium-ion batteries using incremental capacity analysis: a comparative study of single and hybrid deep learning architectures
SUSTAINABLE ENERGY TECHNOLOGIES AND ASSESSMENTS, cilt.93, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 93
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.seta.2026.105340
- Dergi Adı: SUSTAINABLE ENERGY TECHNOLOGIES AND ASSESSMENTS
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, Geobase, INSPEC
- Anahtar Kelimeler: Electric vehicle (EV), Hybrid deep learning, Incremental capacity analysis (ICA), Lithium-ion batteries, State of health (SOH) estimation
- Manisa Celal Bayar Üniversitesi Adresli: Evet
Özet
Accurate state-of-health (SOH) estimation is essential for reliable lifetime management of lithium-ion batteries. However, complex nonlinear electrochemical degradation processes make SOH estimation challenging. This study proposes an electrochemically informed SOH estimation framework integrating incremental capacity analysis (ICA) with single and hybrid deep learning architectures. Voltage, current, temperature, and filtered ICA features are extracted from aging experiments on 2800 mAh lithium-ion cells covering an SOH range of 80 %- 100 %. A systematic comparison is performed using FNN, TCN, GRU, LSTM, Bi-LSTM, and hybrid architectures under identical conditions. In addition to conventional performance metrics, SOH-level error analysis evaluates model behavior across degradation stages, while controlled feature ablation quantifies the incremental contribution of ICA features relative to voltage, current, and temperature inputs. Hybrid architectures outperform single-model approaches. For discharge-profile datasets, TCN + GRU achieves the best performance with an RMSE of 0.73 %, MAE of 0.58 %, and maximum error of 3.08 %, while TCN + GRU + FNN achieves an RMSE of 0.82 % for charge-profile datasets. Ablation results demonstrate consistent reductions in RMSE, MAE, and MAPE with ICA incorporation, confirming that ICA provides additional degradation-sensitive information and that hybrid architectures improve SOH estimation accuracy and stability across the investigated aging stages.