An Explainable Hybrid Opt-GRU-KAN Architecture for Lithium-Ion Battery Health Prediction

State of health (SoH) prediction of lithium-ion batteries is a critical yet challenging task due to the complex, highly non-linear, and time-dependent nature of degradation processes under diverse operating conditions. Variability in usage patterns, environmental factors, and electrochemical dynamics further limits the robustness and generalisation capability of conventional estimation models. This study proposes a hybrid deep learning framework that combines Gated Recurrent Units (GRUs) with Kolmogorov–Arnold Networks (KANs) to address these challenges. GRUs are employed to effectively capture temporal dependencies in sequential battery data, while KANs enhance the model’s ability to learn complex non-linear functional relationships. To improve model transparency and mitigate the black-box nature of deep learning approaches, Explainable Artificial Intelligence (XAI) techniques are integrated to provide interpretable insights into prediction outcomes. Additionally, Optuna-based hyperparameter optimisation is utilised to systematically identify optimal model configurations for enhanced predictive performance. The proposed model is trained using the Toyota Research Institute dataset and validated on real-time electric vehicle battery data collected at Thapar Institute of Engineering and Technology, along with the benchmark CALCE dataset, ensuring robustness across multiple data distributions. Experimental results demonstrate that the hybrid GRU–KAN framework outperforms existing state-of-the-art approaches in terms of accuracy and generalisation. Quantitative evaluation using RMSE, MSE, and MAE shows prediction errors within 1%, indicating high precision in SoH estimation. Module ablation studies further confirm the contribution of each component to the overall model performance.