Lithium-ion batteries power everything from electric vehicles and portable electronics to grid-scale energy storage, thanks to their high energy density, lack of memory effect, and adaptability across temperature ranges. However, repeated charge-discharge cycles cause gradual capacity fade, eventually rendering the battery unusable when it drops below a critical threshold. Accurate prediction of remaining useful life (RUL)—the number of cycles left before this failure point—is essential for proactive battery management, preventing unexpected failures, optimizing replacement schedules, and reducing costs and safety risks in real-world applications.
Traditional RUL prediction methods fall into three categories: physics-based models that simulate internal degradation processes, data-driven approaches that learn patterns from historical data, and hybrid fusions that combine their strengths. While physics-based models offer interpretability, they demand extensive prior knowledge and struggle with complex nonlinear dynamics. Pure data-driven techniques, such as convolutional neural networks (CNNs) for feature extraction or gated recurrent units (GRUs) for time-series forecasting, excel in accuracy when ample high-quality data is available but can accumulate errors over long horizons and lack robustness to noise or limited samples. Hybrid methods address these gaps by integrating probabilistic state estimation, like particle filters (PF), to correct predictions and enhance stability.
A recent study introduces an advanced hybrid framework, the CNN-GRU-PF fusion model, to overcome these limitations. The approach begins by preprocessing battery capacity data using complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) combined with Pearson correlation analysis, effectively decomposing the series into components, reconstructing them to suppress noise, and preserving essential degradation trends. A one-dimensional CNN then extracts high-dimensional spatial features from the processed capacity sequences, while the GRU captures long-term temporal dependencies to generate initial capacity predictions. These predictions serve as observations for the PF, which leverages its strong state estimation capabilities to correct errors and produce optimized outputs. A moving window mechanism iteratively updates the training set by incorporating PF-refined values back into the CNN-GRU model, enabling dynamic adaptation and significantly boosting long-term forecasting performance.
Experimental validation on benchmark NASA datasets, including battery B5, alongside CALCE and custom experimental data, demonstrates the model's superior performance. For battery B5, the CNN-GRU-PF achieves remarkable improvements in prediction accuracy of 87.27% over standalone GRU, 82.88% over PF alone, and 55.43% over the simpler GRU-PF combination. Similar gains appear across other batteries, with enhanced stability even when trained on limited data samples. The iterative updating with the moving window, despite modestly increasing computation time, delivers substantial accuracy gains compared to static versions, underscoring the value of continuous learning in handling evolving degradation patterns.
These advancements promise substantial benefits for battery-dependent technologies. More precise RUL estimates enable better state-of-health monitoring in electric vehicles, extending operational range confidence and preventing abrupt failures that could compromise safety. In energy storage systems, reliable predictions optimize maintenance, reduce downtime, and support efficient integration of renewables. The model's robustness to noise and small datasets makes it practical for diverse operating conditions.
Looking ahead, the CNN-GRU-PF framework holds strong potential for real-time implementation in battery management systems of electric vehicles and grid applications. Future work could validate it on field data from actual vehicles, explore performance under extreme temperatures or abusive conditions, and incorporate additional health indicators like voltage or temperature curves for even greater precision. Extensions to multi-cell packs or different chemistries would broaden applicability, accelerating safer, more sustainable battery utilization.
In essence, this innovative fusion model represents a major stride in battery prognostics by synergistically blending deep learning's pattern recognition with probabilistic filtering's error correction and adaptive training. It delivers unprecedented accuracy and robustness, laying a foundation for smarter battery health management that enhances reliability, longevity, and safety across electrified systems.
Reference
Author: Chunling Wu a b , Chenfeng Xu a b , Liding Wang a b , Juncheng Fu a b , Jinhao Meng c
Title of original paper: Lithium-ion battery remaining useful life prediction based on data-driven and particle filter fusion model
Article link: https://www.sciencedirect.com/science/article/pii/S2773153725000179
Journal: Green Energy and Intelligent Transportation
DOI: 10.1016/j.geits.2025.100267
Affiliations:
a School of Energy and Electrical Engineering, Chang'an University, Xi'an 710064, China
b Shaanxi Key Laboratory of Transportation New Energy Development, Application and Vehicle Energy Saving Technology, Xi'an 710064, China
c School of Electrical Engineering, Xi'an Jiaotong University, Xi'an 710049, China
Green Energy and Intelligent Transportation
Lithium-ion battery remaining useful life prediction based on data-driven and particle filter fusion model
8-Sep-2025