Improved Battery Cycle Life Prediction Using a Hybrid Data-Driven Model Incorporating Linear Support Vector...

Area of Science:

• Battery Technology
• Data Science
• Machine Learning

Background:

• Accurate lithium-ion battery life-time prediction is vital for safety, development, and second-life applications.
• Existing models struggle with early-cycle prediction due to complex, nonlinear battery degradation.
• Early prediction enables proactive management and extends battery operational lifespan.

Purpose of the Study:

• To develop hybrid data-driven models for early-stage lithium-ion battery life-time estimation.
• To address the limitations of current models in predicting degradation at initial usage cycles.
• To validate the models' accuracy using a comprehensive dataset of 124 battery cells.

Main Methods:

• Developed two hybrid models combining linear support vector regression (LSVR) and Gaussian process regression (GPR).
• Utilized a dataset of 124 lithium-ion battery cells with varying lifetimes (150-2300 cycles).
• Evaluated model performance using training and test errors, focusing on early-cycle data (first 100 cycles).

Main Results:

• Achieved low training errors: 1.1% for model A and 1.4% for model B.
• Reported low test errors: 8.3% for model A and 8.2% for model B.
• Maintained prediction error below 10% even when using data from the first 100 cycles.

Conclusions:

• The proposed hybrid models demonstrate high accuracy in predicting lithium-ion battery life-time at early stages.
• The models effectively capture complex degradation patterns, outperforming traditional methods.
• This approach shows significant promise for real-world battery management and second-life applications.