BattVAE-GP predicts battery degradation 100x faster with uncertainty
New AI model simulates 1000+ charge cycles in seconds, not hours.
BattVAE-GP is a hybrid physics-probabilistic framework for modeling lithium-ion battery degradation. It uses a variational autoencoder to compress simulation data into a two-dimensional latent space, then trains a Gaussian process to interpolate degradation dynamics for unseen charging rates, returning uncertainty estimates. Under protocol-level holdout evaluation, the model accurately recovers trajectories for unseen C-rates. By decoding GP-predicted latent states, it produces smooth voltage-capacity curves and uncertainty-aware State of Health estimates.
- Uses a VAE to compress battery degradation simulations into a 2D latent space that captures both cycle progression and charging protocol
- A sparse multitask Gaussian Process interpolates unseen C-rates with uncertainty quantification, reducing compute from hours to seconds
- Achieves accurate recovery of voltage-capacity and State of Health for fully held-out charging conditions
Why It Matters
Faster battery degradation prediction with uncertainty enables cheaper, safer electric vehicle and grid storage development.