Remaining Useful Life Prediction Model of Lithium-Ion Batteries Based on Unscented Kalman Filter-Particle Filter and Enhanced Convolutional Neural Networks

Lithium-ion batteries are widely deployed in electric vehicles and energy storage systems, where accurate prediction of remaining useful life (RUL) is critical for safety and reliability. This paper proposes a novel hybrid prediction model, termed UP-ECNN, which integrates an unscented Kalman filter (UKF)-particle filter (UP) hybrid estimator with an enhanced convolutional neural network (ECNN). First, the UKF-PF module is constructed to mitigate the issue of particle degradation and improve the estimation of nonlinear and non-Gaussian degradation states. Then, the ECNN employs multi-scale skip connections and an attention mechanism to emphasize critical degradation features while capturing both short-term fluctuations and long-term degradation trends. Finally, this hybrid design effectively bridges model-based filtering and data-driven learning, leading to more accurate and reliable RUL prediction. Experimental evaluations on the NASA, CALCE and Oxford battery datasets demonstrate that the proposed UP-ECNN outperforms existing hybrid methods, reducing RMSE and MAE by over 15% on average. These results confirm the superior accuracy, robustness, and practical applicability of the proposed approach for lithium-ion battery health management.