BattVAE-GP Models Battery Degradation with Uncertainty Quantification

Simulating long-term battery degradation is computationally intensive, limiting comprehensive analysis of various operating conditions. Researchers have developed BattVAE-GP, a novel framework that combines physics-based modeling with probabilistic machine learning to create an efficient surrogate model. The system first transforms cycle-resolved degradation data from electrochemical simulations into features, which are then encoded into a 2D latent space using a Variational Autoencoder (VAE). This latent space effectively organizes degradation trajectories by cycle progression and charging protocol. A sparse multitask Gaussian process (GP) then operates within this latent space, interpolating degradation dynamics and providing crucial uncertainty estimates. BattVAE-GP accurately recovers unseen charging rate trajectories and generates smooth voltage-capacity evolution. By propagating the GP's latent posterior through an auxiliary predictor, it also provides uncertainty-aware State of Health (SOH) estimates. This framework offers a computationally efficient and robust tool for predicting battery health under diverse conditions.