XGBoost-Based Remaining Useful Life Estimation Model with Extended Kalman Particle Filter for Lithium-Ion Batteries.

Area of Science:

• Materials Science
• Electrical Engineering
• Data Science

Background:

• Lithium-ion batteries exhibit instability and variable lifetimes, impacting performance and maintenance costs.
• Accurate remaining useful life (RUL) prediction is crucial for dependable, requirement-based maintenance and cost reduction.
• Existing prediction methods struggle to accurately assess battery health and represent uncertainty.

Purpose of the Study:

• To develop a novel particle filter (PF)-based technique for accurate lithium-ion battery RUL estimation.
• To enhance battery dependability and safety through precise state of health (SOH) and RUL prediction.
• To address the limitations of current methods in representing prediction uncertainty.

Main Methods:

• A novel PF-based technique combining a Kalman filter (KF) with a PF for analyzing battery operating data.
• Utilizing extreme gradient boosting (XGBoost) as the core observation model for RUL prediction due to its nonlinear fitting capabilities.
• Conducting life cycle testing to gather precise data for training and validating the RUL prediction model.

Main Results:

• The proposed PF-XGBoost method demonstrated improved accuracy in RUL prediction compared to existing methods.
• The technique effectively maps the relationship between retrieved features and RUL using XGBoost's nonlinear fitting.
• Experimental findings confirm the enhanced accuracy of the suggested technique on a lithium-ion battery cycle life dataset.

Conclusions:

• The developed PF-XGBoost method offers a significant advancement in achieving more accurate RUL prediction for lithium-ion batteries.
• This approach enhances battery safety and dependability by providing reliable estimations of remaining useful life.
• The study highlights the benefit of combining advanced filtering techniques with machine learning for battery health management.