Applications of generative artificial intelligence in battery research: Current status and prospects

With the rapid development of renewable energy and electric vehicles, batteries, as the core components of electrochemical energy storage systems, have become a global focus in both scientific research and industrial sectors due to their critical impact on system efficiency and safety. However, the complex multi-physics reactions within batteries make traditional mathematical models inadequate for comprehensively revealing their mechanisms. The key to solving this problem lies in introducing data-driven approaches, which have laid a solid foundation for battery research and development through extensive accumulation of experimental data and extraction of effective information. Generative artificial intelligence (GAI), leveraging its powerful latent pattern learning and data generation capabilities, has already found widespread applications in protein structure prediction, material inverse design, and data augmentation, demonstrating its broad application prospects. Applying GAI to battery research workflows with diverse battery data resources could provide innovative solutions to challenges in battery research. In this perspective, we introduce the core principles and latest advancements of generative models (GMs), including Generative Adversarial Network (GAN), Variational Auto-Encoder (VAE), and Diffusion Model (DM), which can learn the latent distribution of the input samples to generate new data by sampling from it. Applications of GAI in battery research are then reviewed. For battery materials design, by learning material compositions, structures, and properties, GM can generate novel candidate materials with desired properties through conditional constraints, significantly extending the chemical space to be explored. For electrode microstructure characterization, GM can serve as a bridge for interconversion and integration of different image data, enhance the quality of microscopic characterization, and generate realistic synthetic data. For battery state estimation, GM can perform data augmentation and feature extraction on battery datasets, which benefits the model performance for battery state estimation. Lastly, we discuss the challenges and future development directions in terms of data governance and model design, including data quality and diversity, data standardization and sharing, usability of synthetic data, interpretability of GM, and foundational models for battery research. For the innovation and advancement of battery technology, this perspective offers theoretical references and practical guidelines for implementing GAI as an effective tool in battery research workflows by discussing its status and prospects in this field.