Enhancing Battery SOH Prediction with Butler-Volmer Informed Neural Networks in Data-Scarce Environments
Journal: Energy, 335, 138316 (2025) Role: 1st Author DOI: https://doi.org/10.1016/j.energy.2025.138316
Abstract
This study proposes BVINN (Butler-Volmer Informed Neural Network), a physics-informed machine learning framework that directly incorporates the Butler-Volmer equation into the neural network training process for lithium-ion battery SOH prediction. Experiments on NASA and BIT datasets demonstrate that BVINN achieves high accuracy and physical validity even in data-scarce environments.
Problem & Solution
| Problem | Limitation of Existing Methods | Solution |
|---|---|---|
| Battery stability and lifespan management have become critical due to climate change and EV adoption | Physics-based and data-driven models lack interpretability or generalization, especially with limited data | BVINN incorporates electrochemical principles directly into learning via physics-informed regularization |
Methodology
BVINN incorporates the Butler-Volmer equation as a physics-informed regularization term in the loss function:
- Data Loss (L_data): Minimizes difference between actual data and predictions
- Butler-Volmer Loss (L_BV): Enforces consistency with Butler-Volmer equation-based current-capacity relationship
- Initial Condition Loss (L_IC): Ensures initial values at the first cycle
- Boundary Condition Loss (L_BC): Reflects boundary conditions at the last cycle
- Regeneration Loss (L_regen): Captures temporary capacity recovery (Capacity Regeneration)