BattVAE-GP: Generative Modeling of Long-Horizon Battery Degradation with Uncertainty Quantification

The paper presents BattVAE-GP, a hybrid physics-probabilistic surrogate for long-horizon lithium-ion battery degradation. Cycle-resolved data are generated with a DFN/P2D electrochemical model in PyBaMM, transformed into capacity-aligned voltage and derivative features, and compressed into a two-dimensional VAE latent space. A sparse multitask Gaussian process then models latent degradation dynamics as a function of cycle number and charging C-rate. Protocol-level holdout experiments reportedly recover unseen C-rate trajectories, with uncertainty reflecting the training-data support. Decoding predicted latent states produces smooth voltage-capacity evolution, while Monte Carlo propagation through an auxiliary SOH predictor provides uncertainty-aware health estimates.