Enhanced early prediction of Li-ion battery degradation using multicycle features and an ensemble deep learning model

Achieving high accuracy in the early prediction of Li-ion battery degradation is challenging owing to the nonlinear and dynamic nature of battery aging. This study introduces a GRU-LSTM ensemble model that combines Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) networks to forecast the end-of-life (EoL) of LIBs. The model utilizes features extracted from the voltage and current data during the initial 100 battery cycles and employs a multicycle feature extraction method to enhance computational efficiency without sacrificing predictive accuracy. Hyperparameter tuning via random search and k-fold cross-validation were applied to ensure model robustness. The GRU-LSTM model outperformed the standalone LSTM, GRU, and BiLSTM models across the three input scenarios, achieving the lowest mean absolute percentage error (MAPE) of 5.12 % and root mean squared error (RMSE) of 64.98 cycles using the combined voltage and current features. These results demonstrate the potential of ensemble models to enhance battery health monitoring and predictive maintenance systems.