Generative Modeling of Battery Degradation with Flow Matching and Diffusion Transformers

ICLR 2026 Conference Submission18839 Authors

Keywords: Battery Degradation, State-of-Health Prediction, Flow Matching, Diffusion Transformers, Generative Modeling

TL;DR: We present a scalable generative model for battery degradation that predicts full state-of-health trajectories using flow matching and diffusion transformers.

Abstract: Battery degradation remains a critical challenge in the pursuit of green technologies and sustainable energy solutions. Despite significant research efforts, predicting battery cycle life accurately remains difficult due to the complex interplay of aging and cycling behaviors. To address this challenge, we introduce FlowBatt, a general-purpose model for battery degradation prediction and synthesis trained via flow matching. FlowBatt leverages a scalable diffusion transformer (DiT) backbone, enabling high expressivity and scalability. The model operates as a probabilistic predictor of entire cycle life trajectories and as a generative model capable of synthesizing realistic degradation curves for data augmentation. We demonstrate the advantages of flow-based generative approaches by comparing models trained with flow matching, diffusion processes, and supervised learning. FlowBatt achieves results that are comparable to or better than state-of-the-art performance for the remaining useful life prediction task and provides accurate and generalizable state-of-health predictions while capturing uncertainty in aging dynamics. Beyond prediction accuracy, this work advances the development of foundational and scalable models for battery degradation.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 18839