High-Throughput Screening of Ni-Rich Layered Cathode Materials for Sodium-Ion Batteries Based on Machine Learning

Ni-rich layered cathode materials have attracted widespread attention due to their high capacity and cycle life for sodium-ion batteries (SIBs). Screening stable Ni-rich layered cathode materials is extremely important for improving their electrochemical performance. In this paper, a transfer learning (TL) combined with deep neural networks (DNN) method is developed to predict the average voltage and specific capacity of Ni-rich layered cathode materials, and to screen doped cathode materials with excellent electrochemical performance. First, two training datasets from Materials Project (MP) are constructed for initial training and retraining of DNN model, respectively. 2673 potential cathode materials are obtained by using element substitution as test dataset for DNN model. The best predicted results by using initial trained DNN models showed the mean absolute errors of 0.44V and 27.36mAgh-1 for voltage and specific capacity, respectively. Then, a TL method is integrated into our DNN model for retraining. The performance of retrained DNN model is significantly improved to 0.25V for voltage and 60.23mAgh-1 for specific capacity, compared with the previous DNN model. Finally, 391 cathode materials are identified for Ni-rich SIBs with average voltage > 3.8V and specific capacity > 240 mAgh-1, most of which are Co-free materials. This work provides a fast and effective method for high-throughput screening of Ni-rich cathode materials with high average voltage and specific capacity.