Feature and LSTM-Based Capacity Prediction for Lithium-Ion Batteries: Application of Linear Regression and LSTM Models, with Future Expansion to Fuel Cell Datasets and Optimization

Akkaya S., Uyar M.

9th International Hydrogen Technologies Congress, İzmir, Türkiye, 25 - 28 Mayıs 2025, ss.915-918, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Basıldığı Şehir: İzmir
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.915-918
  • Bursa Uludağ Üniversitesi Adresli: Evet

Özet

Lithium-ion battery capacity forecasting is critical to optimizing energy storage devices, and equivalent forecasts for fuel cells are needed to attain power generation efficiency and extended system lives. The current research used discharge data with lithium-ion battery capacity values to create four single time-series data sets, of which 10 features per series were created, resulting in 40 feature matrices. Both linear regression and Long Short-Term Memory (LSTM) models were used to predict battery capacity. The linear regression model worked with an RMSE of 0.0058, and the LSTM-based model was compared based on RMSE and loss function metrics. The results indicate the possibility of applying these methods to predict capacity accurately. The proposed approach sets a foundation for subsequent optimization and adjustment toward fuel cell data for more stable prediction and better control over the energy system in future work.