EXTENDED ABSTRACT: The Graph neural network is a machine learning method developed in recent years, which can transform data into graphs and extract information such as connections from the graphs, thus possessing strong capabilities in feature extraction. In this study, we developed a crystal graph neural network (CGCNN) based machine learning model predict the voltages for zinc metal cathode materials, using basic principles of material thermodynamics and materials data from the Materials Project and AFLOW databases. The data from both databases was cross validated, and more than 80 highvoltage, high speciffc energy zinc metal cathode materials were predicted without adding any experimental data. At the same time, hundreds of metal alloy anode materials for various batteries such as Li, Na, K, Ca, Mg, Zn, and Al were predicted and compared with experimental data, and the model performed excellently. We hope to spur further interest in expanding the applications of machine learning methods for accelerating the design of high performance battery materials.
Prof. Zijian Hong is currently a tenure-track faculty member at School of Materials Science and Engineering, Zhejiang University. He obtained his Ph.D. degree from Penn State University in 2017 under the supervision of Prof. Long-Qing Chen. Then he worked as a postdoc research fellow at Carnegie Mellon University for 3 years. In 2020, he was appointed as a 100 Talent Fellow at the School of Materials Science and Engineering, Zhejiang University. His research primarily focuses on computational materials science, particularly mesoscale simulations and machine learning. He has published over 50 SCI papers, which have received 3,000 citations, including three in Nature and four in Nature Materials. He has also delivered invited talks at more than ten universities and conferences, including UC Berkeley and the University of Maryland.