Alexander Popp auf IEEE International Conference on Energy Technologies for Future Grids (ETFG 2025)
Folgendes Thema wurde im Rahmen der i-COSTE 2025 Konferenz präsentiert:
Cross-Dataset Transferability Analysis of Neural Networks for Battery SOH Prediction Using Synthetic and Experimental Data
A. Popp, U. Spaeth, S. Qaadan, H. Fechtner, and B. Schmuelling
This study investigates the effectiveness of artificial neural networks for lithium-ion battery State of Health estimation using both synthetic and experimental datasets. The research evaluates model performance across different datasets and examines the transferability of models trained on synthetic data to real-world experimental conditions. Results demonstrate excellent predictive accuracy for models trained and tested on matching datasets, with synthetic models achieving RMSE values around 0.001 and experimental models reaching RMSE of 0.023. However, significant limitations emerge in cross-dataset transferability, with synthetic models showing poor performance on experimental data (RMSE: 4.965). The findings highlight the critical need for hybrid modeling approaches that effectively combine synthetic and experimental data to bridge the gap between physics-based simulations and real-world battery behavior in practical applications.
Wir freuen uns sehr über die erfolgreiche Teilnahme und die Möglichkeit, unsere Forschung auf internationaler Bühne präsentieren zu können.
Herzlichen Glückwunsch an unseren Kollegen zu diesem wertvollen Beitrag!
