Comparative Analysis of Hybrid CNN–LSTM and CNN–TCN Models for State of Charge Estimation in Li-ion Batteries
5th International Conference on Computer Engineering, Technologies and Applications (CETA 2026), Antalya, Türkiye, 27 - 29 Nisan 2026, sa.39, ss.1-5, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Basıldığı Şehir: Antalya
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1-5
- Ankara Üniversitesi Adresli: Evet
Özet
Accurate estimation of the State of Charge (SoC) is essential for ensuring the reliability of battery management systems in electric vehicles. However, the nonlinear electrochemical behavior of Li-ion batteries and their sensitivity to variables such as temperature and current make high-accuracy SoC estimation challenging. In this study, a hybrid deep learning–based approach is proposed for SoC estimation, and single models including Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Temporal Convolutional Network (TCN), together with hybrid models (CNN-LSTM, CNN-TCN, and TCN-LSTM), are comparatively evaluated. Multivariate time-series data consisting of battery voltage, temperature, charge rate, and cycle index were processed using the sliding window method, and the models were trained using an 80% training and 20% testing data split. Experimental results show that the CNN–LSTM hybrid model achieved the highest prediction accuracy with MAE = 1.093, RMSE = 2.405, and R2 = 0.993. In addition, the single CNN model achieved R2 = 0.990, while the CNN–TCN model obtained R2 = 0.980. The findings demonstrate that hybrid architectures combining local feature extraction with temporal modeling provide an effective and reliable solution for SoC estimation.