Early Remaining Useful Life Prediction of Lithium-Ion Batteries Based on a Hybrid Machine Learning Method with Time...

Area of Science:

• Battery health monitoring
• Prognostics and Health Management (PHM)
• New Energy Vehicles (NEVs)

Background:

• Accurate Remaining Useful Life (RUL) prediction is vital for NEV reliability.
• Challenges include noise, capacity regeneration, and data scarcity in early prognostics.
• Existing methods struggle with limited early-stage battery data.

Purpose of the Study:

• To develop a robust hybrid framework for early and accurate battery RUL prediction.
• To address data scarcity and noise contamination in battery prognostics.
• To enhance the reliability of battery management systems in NEVs.

Main Methods:

• Signal decomposition using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN).
• Time series augmentation with a Transformer-enhanced Generative Adversarial Network (HyT-GAN).
• Deep forecasting using a CNN-BiGRU predictor optimized by the Dung Beetle Optimizer (DBO).

Main Results:

• Accurate early-stage RUL prediction achieved using only 20% of historical data.
• High prediction accuracy demonstrated with R2 values from 0.9643 to 0.9972.
• Excellent performance metrics (RMSE/MAE < 0.0296/0.0198) on benchmark datasets (NASA, CALCE).

Conclusions:

• The proposed hybrid framework provides reliable RUL estimates under data-limited and noisy conditions.
• The method significantly improves early prognostics for battery health management.
• This approach enhances the operational safety and efficiency of new energy vehicles.