Eğitim-Test Bölme Oranının LSTM Tabanlı Batarya Şarj Durumu Tahmin Modeli Performansı Üzerindeki Etkileri: Karşılaştırmalı Bir Çalışma

Öz

Bağlam—Elektrikli araç (EV) bataryalarının yönetiminde batarya şarj durumu (SOC) tahmini kritik bir parametre olarak kabul edilir. Doğru ve güvenilir SOC tahmini, batarya şarj ve deşarj süreçlerinin etkin kontrolüne, enerji yönetiminin optimizasyonuna ve batarya ömrünün uzatılmasına doğrudan katkıda bulunur. Hassas SOC tahmini; EV'lerin daha güvenilir, sürdürülebilir ve daha yüksek verimlilikle çalıştığını gösterir. Bu nedenle, SOC tahmini batarya yönetim sistemlerinin (BMS) temel bileşenlerinden biri olarak kabul edilir. Amaç—Bu çalışmada, SOC tahmini yapmak için derin öğrenmeye dayalı bir yaklaşım olan uzun kısa süreli bellek (LSTM) modeli seçilmiştir. Yöntem—Model uygulamasında kullanılan veri seti, modelin öğrenme ve doğrulama performansını değerlendirmek için farklı oranlarda eğitim ve test veri setlerine ayrılmıştır. Eğitim ve test veri oranlarının model performansı üzerindeki etkisini analiz etmek için ilk aşamada dört farklı senaryo oluşturulmuştur. Bu senaryolarda, eğitim-test veri bölme oranları sırasıyla %60-%40, %70-%30, %80-%20 ve %90-%10 olarak belirlenmiştir. İkinci aşamada, bu dört farklı veri bölme senaryosu ayrı ayrı LSTM modeline uygulanmış ve modelin tahmin performansı her durumda ayrıntılı olarak incelenmiştir. Son aşamada, elde edilen sonuçlar, model performansını niceliksel olarak değerlendirmek için çeşitli hata ölçütleri kullanılarak analiz edilmiştir. Bu bağlamda, ortalama mutlak hata (MAE), ortalama kare hata (MSE), kök ortalama kare hata (RMSE) ve belirleme katsayısı (R²) değerleri hesaplanmış ve karşılaştırmalı olarak değerlendirilmiştir. Bulgular— Analizler sonucunda, model parametreleri sabit tutulduğunda bile, eğitim ve test veri oranlarının SOC tahmin performansı üzerinde önemli bir etkisi olduğu açıkça gözlemlenmiştir. Bulgulara göre, eğitim ve test veri oranı %80-%20 olarak ayarlandığında en düşük hata değerleri elde edilmiştir. Bu senaryoda, MSE değeri %1,3305, MAE değeri %0,8003, RMSE değeri %1,1535 ve R² değeri %93,53 olarak hesaplanmıştır. Buna karşılık, en yüksek hata değerlerinin %60-%40 eğitim-test oranında meydana geldiği belirlenmiştir. Bu durumda, MSE %4,9628, MAE %1,7788, RMSE %2,2277 ve R² %73,27 olarak bulunmuştur. Sonuç— Ayrıca, değerlendirmeler, eğitim verisi oranı azaldıkça modelin eğitim süresinin de kısaldığını göstermiştir. Bununla birlikte, bu durum modelin tahmin doğruluğunu olumsuz etkilemiş ve daha düşük performansa yol açmıştır. Bu bulgular, eğitim verisi miktarının model performansı üzerindeki kritik rolünü açıkça göstermektedir.

Anahtar Kelimeler

• Elektrikli araçlar
• Batarya şarj durumu tahmini
• Uzun kısa süreli bellek
• Eğitim-test bölme oranı

Destekleyen Kurum

Türkiye Bilimsel ve Teknolojik Araştırma Kurumu (TÜBİTAK)

Etik Beyan

Hazırlanan makale için etik kurulundan izin alınmasına gerek yoktur.

Teşekkür

Bu çalışma, Türkiye Bilimsel ve Teknolojik Araştırma Kurumu (TÜBİTAK) tarafından 2211 numaralı proje kapsamında desteklenmiştir.

Effects of training-test splitting ratio on the performance of LSTM-based battery state of charge estimation model: a comparative study

Öz

Context—Battery state of charge (SOC) estimation is considered a critical parameter in the management of electric vehicle (EV) batteries. Accurate and reliable SOC estimation directly contributes to the effective control of battery charging and discharging processes, optimization of energy management, and extension of battery life. Furthermore, precise SOC estimation demonstrates that EV operate more reliably, sustainably, and with higher efficiency. Therefore, SOC estimation is considered one of the fundamental components of battery management systems (BMS). Objective— In this study, a long short-term memory (LSTM) model, a deep learning-based approach, was chosen to perform SOC estimation. Method— The dataset used in the model implementation was divided into training and test datasets with different ratios to evaluate the model's learning and validation performance. To analyze the effect of training and test data ratios on model performance, four different scenarios were created in the first stage. In these scenarios, the training-test split ratios were determined as 60%-40%, 70%-30%, 80%-20%, and 90%-10%, respectively. In the second stage, these four different data splitting scenarios were applied separately to the LSTM model, and the model's estimation performance was examined in detail in each case. In the final stage, the obtained results were analyzed using various error metrics to quantitatively evaluate the model performance. In this context, mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R²) values were calculated and evaluated comparatively. Results— As a result of the analyses, it was clearly observed that even when the model parameters were kept constant, the training and test data ratios had a significant effect on the SOC estimation performance. According to the findings, the lowest error values were obtained when the training and test data ratio was set at 80%-20%. In this scenario, the MSE value was calculated as 1.3305%, the MAE value as 0.8003%, the RMSE value as 1.1535%, and the R² value as 93.53%. In contrast, the highest error values were determined to have occurred at a training-test ratio of 60%-40%. In this case, the MSE was found to be 4.9628%, MAE 1.7788%, RMSE 2.2277%, and R² 73.27%. Conclusion— Furthermore, the evaluations showed that as the training data ratio decreased, the model's training time also shortened. However, this negatively impacted the model's estimation accuracy and resulted in lower performance. These findings clearly demonstrate the crucial role of the amount of training data on model performance.

Anahtar Kelimeler

• Electric vehicles
• Battery state of charge estimation
• Long short term memory
• Training-test splitting ratio

Destekleyen Kurum

Scientific and Technical Research Council of Türkiye (TUBITAK)

Etik Beyan

There is no need to obtain permission from the ethics committee for the article prepared.

Teşekkür

This study was supported in the context of 2211 project by Scientific and Technical Research Council of Türkiye (TUBITAK).

Kaynakça

1. [1] M. Waseem, G. S. Lakshmi, M. Amir, M. Ahmad, M. Suhaib, “Advancement in battery health monitoring methods for electric vehicles: Battery modelling, state estimation, and internet-of-things based methods”, Journal of Power Sources, 633, 236414, 2025. https://doi.org/10.1016/j.jpowsour.2025.236414.
2. [2] F. Liu, D. Yu, C. Shao, X. H. Liu, W. X. Su, “A review of multi-state joint estimation for lithium-ion battery: Research status and suggestions”, Journal of Energy Storage, 73, 109071, 2023. https://doi.org/10.1016/j.est.2023.109071.
3. [3] L. Efe, A. Tabak, “State estimation for electric vehicles with deep learning-a survey”, Electric Power Systems Research, 255, 112800, 2026. https://doi.org/10.1016/j.epsr.2026.112800.
4. [4] Z. Ren, C. Du, “A review of machine learning state-of-charge and state-of-health estimation algorithms for lithium-ion batteries”, Energy Reports, 9, 2993-3021, 2023. https://doi.org/10.1016/j.egyr.2023.01.108.
5. [5] E. Chemali, P. J. Kollmeyer, M. Preindl, R. Ahmed, A. Emadi, “Long short-term memory networks for accurate state-of-charge estimation of Li-ion batteries”, IEEE Transactions on Industrial Electronics, 65(8), 6730-6739, 2018. https://doi.org/10.1109/TIE.2017.2787586.
6. [6] F. F. Yang, X. B. Song, F. Xu, K. L. Tsui, “State-of-charge estimation of lithium-ion batteries via long short-term memory network”, IEEE Access, 7, 53792-53799, 2019. https://doi.org/10.1109/ACCESS.2019.2912803.
7. [7] X. B. Song, F. F. Yang, D. Wang, K. L. Tsui, “Combined CNN-LSTM network for state-of-charge estimation of lithium-ion batteries”, IEEE Access, 7, 88894-88902, 2019. https://doi.org/10.1109/ACCESS.2019.2926517.
8. [8] M. Wei, M. Ye, J. B. Li, Q. Wang, X. X. Xu, “State of charge estimation of lithium-ion batteries using LSTM and NARX neural networks”, IEEE Access, 8, 189236-189245, 2020. https://doi.org/10.1109/ACCESS.2020.3031340.

9. [9] Y. Tian, R. Lai, X. Li, L. Xiang, J. Tian, “A combined method for state-of-charge estimation for lithium-ion batteries using a long short-term memory network and an adaptive cubature Kalman filter”, Applied Energy, 265, 114789, 2020. https://doi.org/10.1016/j.apenergy.2020.114789.
10. [10] F. F. Yang, S. H. Zhang, W. H. Li, Q. Miao, “State-of-charge estimation of lithium-ion batteries using LSTM and UKF”, Energy, 201, 117664, 2020. https://doi.org/10.1016/j.energy.2020.117664.
11. [11] M. Fasahat, M. Manthouri, “State of charge estimation of lithium-ion batteries using hybrid autoencoder and Long Short Term Memory neural networks”, Journal of Power Sources, 469, 228375, 2020. https://doi.org/10.1016/j.jpowsour.2020.228375.
12. [12] D. W. Chung, J. H. Ko, K. Y. Yoon, “State-of-charge estimation of lithium-ion batteries using LSTM deep learning method”, Journal of Electrical Engineering & Technology, 17(3), 1931-1945, 2022. https://doi.org/10.1007/s42835-021-00954-8.
13. [13] X. H. Zhang, Y. Huang, Z. W. Zhang, H. P. Lin, Y. Zeng, M. Y. Gao, “A hybrid method for state-of-charge estimation for lithium-ion batteries using a long short-term memory network combined with attention and a Kalman filter”, Energies, 15(18), 6745, 2022. https://doi.org/10.3390/en15186745.
14. [14] C. S. Hu, B. H. Li, L. Ma, F. J. Cheng, “State-of-charge estimation for lithium-ion batteries of electric vehicle based on sensor random error compensation”, Journal of Energy Storage, 55, 105537, 2022. https://doi.org/10.1016/j.est.2022.105537.
15. [15] C. N. Van, M. D. Ngo, C. D. Duc, L. Q. Thao, S. J. Ahn, “State-of-Charge Estimation of Lithium–Ion Battery integrated in Electrical Vehicle Using a Long Short-Term Memory Network”, IEEE Access, 12, 165472-165481, 2024. https://doi.org/10.1109/ACCESS.2024.3493141.
16. [16] A. Praisan, S. Polmai, S. Kittiratsatcha, “A Standalone Real-Time State of Charge Estimator Using the Long Short-Term Memory Artificial Neural Network”, International Review of Electrical Engineering, 19(6), 482–496, 2024. https://doi.org/10.15866/iree.v19i6.25603.
17. [17] R. H. Wong, D. Sooriamoorthy, A. Manoharan, N. Binti Sariff, Z. Hilmi Ismail, “Balancing accuracy and efficiency: a homogeneous ensemble approach for lithium-ion battery state of charge estimation in electric vehicles”, Neural Computing and Applications, 36(30), 19157-19171, 2024. https://doi.org/10.1007/s00521-024-10210-5.
18. [18] X. Q. Chai, S. H. Li, F. W. Liang, “A novel battery SOC estimation method based on random search optimized LSTM neural network”, Energy, 306, 132583, 2024. https://doi.org/10.1016/j.energy.2024.132583.
19. [19] H. Q. Xu, Q. Xu, F. Duanmu, J. Y. Shen, L. Jin, B. Gou, F. Wu, W. Zhang, “State-of-Charge Estimation of Lithium-Ion Batteries Based on EKF Integrated With PSO-LSTM for Electric Vehicles”, IEEE Transactions on Transportation Electrification, 11(1), 2311-2321, 2024. https://doi.org/10.1109/TTE.2024.3421260.
20. [20] S. Kumar, A. B. Choudhury, H. S. Bhattacharyya, C. K. Chanda, “Comparison of different data-driven methods for estimating the state of charge of lithium-ion batteries”, Energy Sources Part A - Recovery, Utilization, and Environmental Effects, 47(1), 2564-2583, 2025. https://doi.org/10.1080/15567036.2025.2452248.
21. [21] L. Pu, C. Wang, “Combined State-of-Charge Estimation Method for Lithium-Ion Batteries Using Long Short-Term Memory Network and Unscented Kalman Filter”, Energies, 18(5), 1106, 2025. https://doi.org/10.3390/en18051106.
22. [22] Y. R. Zou, S. L. Wang, W. Cao, N. Hai, C. Fernandez, “Enhanced transformer encoder long short-term memory hybrid neural network for multiple temperature state of charge estimation of lithium-ion batteries”, Journal of Power Sources, 632, 236411, 2025. https://doi.org/10.1016/j.jpowsour.2025.236411.
23. [23] A. Elachhab, E. M. Laadissi, A. Tabine, A. Hajjaji, “Deep learning and data augmentation for robust battery state of charge estimation in electric vehicles”, Electrical Engineering, 107(6), 7313-7327, 2025. https://doi.org/10.1007/s00202-024-02917-4.
24. [24] J. Ahn, Y. Lee, B. Han, S. Lee, Y. Kim, D. Chung, J. Jeon, “A highly effective and robust structure-based LSTM with feature-vector tuning framework for high-accuracy SOC estimation in EV”, Energy, 325, 136134, 2025. https://doi.org/10.1016/j.energy.2025.136134.
25. [25] R. Carrera, L. Quiroz, C. Guevara, P. Acosta-Vargas, “State-of-Charge Estimation of Medium-and High-Voltage Batteries Using LSTM Neural Networks Optimized with Genetic Algorithms”, Sensors, 25(15), 4632, 2025. https://doi.org/10.3390/s25154632.
26. [26] B. J. Yu, G. Wang, E. N. Zhu, S. Y. Yao, Y. Zhou, “Predicting lithium-ion battery state of charge with long short-term memory network enhanced extended Kalman filter”, Journal of Energy Storage, 132, 117849, 2025. https://doi.org/10.1016/j.est.2025.117849.
27. [27] E. D. Bobobee, S. L. Wang, P. Takyi-Aninakwa, G. C. Liu, E. Koukoyi, “State of charge estimation of lithium-ion batteries using improved multi-attention long short-term memory extended Kalman filtering model”, Engineering Applications of Artificial Intelligence, 158, 111526, 2025. https://doi.org/10.1016/j.engappai.2025.111526.
28. [28] J. Schaeffer, E. Lenz, D. Gulla, M. Z. Bazant, R. D. Braatz, R. Findeisen, “Gaussian process-based online health monitoring and fault analysis of lithium-ion battery systems from field data”, Cell Reports Physical Science, 5(11), 102258, 2024. https://doi.org/10.1016/j.xcrp.2024.102258.
29. [29] S. Hochreiter, J. Schmidhuber, “Long short-term memory”, Neural computation, 9(8), 1735-1780, 1997. https://doi.org/10.1162/neco.1997.9.8.1735.
30. [30] S. Siami-Namini, N. Tavakoli, A. S. Namin, “The performance of LSTM and BiLSTM in forecasting time series”, 2019 IEEE International Conference on Big Data, Los Angeles, CA, USA, 09-12 December 2019. https://doi.org/10.1109/BigData47090.2019.9005997.
31. [31] Massachusetts Institute of Technology, “Lithium-Ion Battery Field Data: 28 LFP battery systems with 8 cells in series, up to 5 years of operation”, https://zenodo.org/records/13715694 (14.05.2026).