Akademik Veri Yönetim Sistemi
ICSAS 6TH INTERNATIONAL CONFERENCE ON APPLIED SCIENCES, İzmir, Türkiye, 6 - 08 Şubat 2026, ss.388-399, (Tam Metin Bildiri)
Lithium-ion batteries are widely used as critical energy storage components in many applications, particularly in electric vehicles and portable electronic devices. Accurate prediction of their remaining useful life (RUL) is crucial for system reliability and predictive maintenance. In this study, the predictive performance of LSTM, GRU, and BiLSTM deep learning models for capacity degradation is compared using the Li-ion battery aging dataset from the NASA Prognostics Center of Excellence (PCoE). The model hyperparameters (number of hidden-layer neurons, lookback window size, and learning rate) are optimized using the Random Search method. The experimental studies are conducted on three batteries (B0005, B0006, B0007), and model performances are evaluated using RMSE, MAE, and MAPE metrics. The results show that the GRU model achieves the lowest MAPE and performs better than LSTM and BiLSTM. The GRU model for the B0006 battery achieves the lowest RMSE of 0.0155. These findings reveal that the GRU architecture is a promising alternative for battery prognostics, offering both computational efficiency and prediction accuracy.