Prediction on Discharging Properties of Nickel–Manganese Materials for High-Performance Sodium-Ion Batteries via Machine Learning Methods

Abstract

Understanding the cyclic discharge feature of oxide cathodes, such as sodium-based nickel–manganese (NMMn), determines the future applications of sodium-ion batteries. Machine learning methods, including gradient-boosting models and random forest (RF) machine, are applied to an experimental dataset of these NMMn materials. Herein, gradient-boosting models achieve a better performance than RF machine in predicting the discharge properties of these materials. The results indicate that the dopant content ratio, sodium content, and nickel content play important roles in the initial discharge capacities (IC) and 50th cycle end discharge capacities (EC) of these materials. NMMn cathodes with a specific sodium content (0.75 < x < 1.25), a dopant content (x < 0.2), and a nickel content (x < 0.4) are more likely to possess high ICs and ECs. Unlike the cathode for lithium-ion batteries, herein, nickel content in NMMn affects more on 50th cycle EC. These results offer new guidelines to design high-performance cathodes for sodium-ion batteries.