Realistic data-driven deep learning framework with multi-objective optimization for lithium-ion battery state estimation
Journal of Energy Storage, cilt.183, 2027 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 183
- Basım Tarihi: 2027
- Doi Numarası: 10.1016/j.est.2026.125433
- Dergi Adı: Journal of Energy Storage
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC
- Anahtar Kelimeler: MOGA-BiLSTM-SA, Self-attention mechanism, SoC estimation, Vehicle-level simulation data
- Yozgat Bozok Üniversitesi Adresli: Evet
Özet
State-of-charge (SoC) estimation is essential for the safe and efficient operation of lithium-ion batteries in electric vehicles (EVs). However, accurate estimation remains challenging due to nonlinear battery behavior, temperature variations, and aging effects. We introduce a hybrid Bidirectional Long Short-Term Memory (BiLSTM) self-attention (SA) framework with hyperparameters optimized via a Multi-Objective Genetic Algorithm (MOGA) for enhanced performance. The proposed method is evaluated using the publicly available Samsung INR21700-30T lithium-ion battery dataset, together with dSPACE-based battery electric vehicle real SoC simulation experiment data, obtained in real-time. Several hours of high-fidelity driving data are generated using the dSPACE simulation platform based on the standard Federal Test Procedure 75 (FTP-75) drive cycle, representing realistic vehicle operating conditions. The study performs a comprehensive comparative evaluation of SoC estimation accuracy over a broad temperature, ranging 40 °C to −20 °C, demonstrating its effectiveness under varying ambient temperature conditions. In this context, unlike many studies that rely solely on laboratory-obtained SoC data, we leverage high-fidelity vehicle data, capturing the stochastic nature of actual driving behavior and enhancing the temporal modeling capabilities of MOGA-based BiLSTM-SA framework for high-accuracy SoC estimation in advanced BMS applications. Under the UDDS cycle, the MOGA-BiLSTM-SA model consistently achieves the lowest RMSE value at most of the tested temperatures and the lowest MAE value across all temperature values, demonstrating robust predictions under structured urban driving conditions. Under the FTP-75 cycle, the MOGA-BiLSTM-SA model remains competitive across all tested temperatures as the best-performing architecture varies with temperature.