Machine Learning-Assisted KCl-CaCl2-LiCl Electrolyte Design for Low-Temperature, High-Performance Calcium-Based Liquid Metal Batteries.

Calcium-based liquid metal batteries are promising for large-scale energy storage due to calcium abundance and low cost, yet their practical applications are impeded by high operating temperatures, severe self-discharge, limited coulombic efficiency, and rapid capacity fading. Here, we develop a machine learning (ML)-assisted optimization framework, integrating data-driven analysis, ML prediction, and experimental verification to design a high-performance ternary molten-salt electrolyte. Through multi-parameter evaluation of thermodynamic stability, melting behavior, density, and cost, KCl was identified as an optimal third component for CaCl2-LiCl based systems. A multidimensional descriptor-performance dataset was constructed to develop a random forest model for precise electrolyte composition optimization. Guided by this model, the KCl-CaCl2-LiCl electrolyte (13:35:52 mol%) was experimentally verified to enable stable operation at 480°C, delivering a coulombic efficiency >99.5%, an ultralow self-discharge current density of 0.31 mA cm-2, >91% capacity retention after 100 cycles, while maintaining a low material cost of 0.81$ kg-1. This optimized ternary electrolyte suppresses calcium dissolution through cooperative multi-cation effects, significantly improving low-temperature electrochemical performance and cycling stability. This work not only provides a viable pathway toward practical Ca-based LMBs but also establishes a generalizable ML-assisted paradigm for accelerated electrolyte design in advanced electrochemical energy storage.