A lotus-optimized Radial basis function framework for explainable and energy-efficient battery health prediction in electric vehicles

Lithium-ion batteries (LiBs) are central to modern electric mobility, yet accurate health prediction remains challenging due to nonlinear degradation, thermal variability, and noisy operational data. This study presents a novel hybrid framework—the Lotus-based Radial Basis Function (LbRBF) model—which integrates the bio-inspired Lotus Optimization Algorithm (LOA) with Radial Basis Function Neural Networks (RBFNNs) for intelligent, adaptive, and computationally efficient battery health prediction. Trained on real-world NASA and Oxford EV battery datasets, LbRBF achieved an R2 of 0.988, RMSE of 9.90%, and MAE of 0.49%, outperforming state-of-the-art models such as LSTM, CNN, and SVM by up to 12.5% in prediction accuracy. The model demonstrates high computational efficiency, achieving 730 inferences/s with only 3.6 × 105 FLOPs, indicating suitability for low-latency applications. Although experimental validation was conducted on an Intel i7 CPU and NVIDIA RTX 3060 GPU, the low computational complexity suggests promising adaptability to resource-constrained embedded BMS platforms, pending dedicated hardware-level validation. Additionally, SHAP-based explainability provides insights into dominant degradation factors, including temperature and overcharge rate, improving model transparency. By combining high predictive accuracy, energy-efficient operation, and interpretability, the proposed LbRBF framework offers a scalable solution for next-generation electric vehicles and smart energy storage systems, enabling proactive battery management, optimized charging strategies, and extended battery lifespan.