Accurate and efficient prediction on the formation energy and potential profiles of sodium vanadium oxyfluorophosphate by machine learning

As the demand for sustainable energy solutions increases, sodium-ion batteries (SIBs) have emerged as a promising alternative to lithium-ion batteries due to the abundance and low cost of sodium resources. However, a key challenge in advancing SIB technology is the development of electrode materials with stable voltage outputs. NASICON-type vanadium phosphate cathodes, particularly sodium vanadium oxyfluorophosphates (NVPFO), have shown great potential but still face limitations in terms of slow Na⁺ diffusion kinetics. Traditional experimental methods for optimizing material compositions are time-consuming and often unreliable, while Density Functional Theory (DFT) is computationally expensive because it involves iterative quantum mechanical calculations whose cost increases rapidly with system size, making it less practical for high-throughput studies of large or complex materials. To address these challenges, this study employs machine learning (ML) to predict formation energies and voltage profiles of NVPFO cathodes, which are crucial for understanding the electrode's thermodynamic stability and electrochemical performance. Leveraging data from the Materials Project database, nine ML models were evaluated with XGBoost achieving the highest accuracy (R² = 0.9593 after feature selection). Feature engineering and SHAP analysis identified the key descriptors that govern material performance. A comparative analysis with DFT confirmed that ML predictions were not only accurate but also computationally efficient, offering a significantly faster alternative. Discharge voltage curves based on ML model is also closely with experimental profiles, but 50000 times faster than that based on DFT calculation. This study demonstrates that ML can accelerate the design and optimization of high-performance NVPFO materials for SIBs, offering a scalable and efficient alternative to traditional methods.