Application and feasibility analysis of random Forest models in safety prediction for transition metal-doped Na3V2(PO4)3 sodium-ion batteries.

With the increasing depletion of traditional energy resources and the continuous rise in usage costs, the development of new energy storage technologies has become a global focus. Although lithium-ion batteries possess excellent electrochemical performance and a well-established industrial infrastructure, they are prone to thermal runaway and pose significant safety risks. As a potential alternative to lithium, sodium-ion batteries demonstrate promising commercial prospects due to their low cost, similar electrochemical properties, and the fact that they are more resistant to thermal runaway and exhibit less violent reactions. Among the numerous cathode materials for sodium-ion batteries, Na3V2(PO4)3 (NVP) has attracted significant attention due to its high energy density, structural stability and excellent thermal safety; however, its inherently low electronic and ionic conductivity limits its practical application. To address this, this study modifies NVP by introducing different transition metal elements to regulate the unit cell volume, broaden the ionic migration pathways, and further enhance its electrochemical performance and thermal safety. Electrochemical test results indicate that the optimal doped sample achieves a half-cell/full-cell capacity of up to 122.3 mAh g-1; infrared spectroscopy reveals minimal temperature variation during charge-discharge cycles. Accelerated Calorimetry (ARC), Thermogravimetric Analysis (TG) and Differential Scanning Calorimetry (DSC) tests further confirmed that element doping significantly improves the thermal stability of NVP, increases heat absorption and reduces the risk of thermal runaway. Building on this, a Random Forest (RF) machine learning model was developed to systematically analyze thermal runaway data and predict safety performance under different doping conditions. Model evaluation results indicate that the RF model possesses high predictive accuracy and good generalisation ability, enabling it to effectively identify key features influencing safety. Integrating experimental and modelling results, the combination of transition metal element doping modification and machine learning prediction methods provides a new technical pathway and theoretical basis for the design and feasibility assessment of high-safety sodium-ion batteries.