Machine learning for battery research

Abstract

Batteries are vital energy storage carriers in industry and in our daily life. There is continued interest in the developments of batteries with excellent service performance and safety. Traditional trial-and-error experimental approaches have the limitations of high-cost and low-efficiency. Atomistic computational simulations are relatively expensive and take long time to screen massive materials. The rapid development of machine learning (ML) has brought innovations in many fields and has also changed the paradigm of the battery research. Numerous ML applications have emerged in the battery community, such as novel materials discovery, property prediction, and characterization. In this review, we introduced the workflow of ML, where the task, data, feature engineering, and evaluation were involved. Several typical ML models used in batteries were highlighted. In addition, we summarized the applications of ML for the discovery of novel materials, and for property and battery state prediction. The challenges for the application of ML in batteries were also discussed.