AI-Informed Solvation Engineering for Thermogalvanic Electrolytes with High Thermopower

Thermogalvanic cells utilize redox couple ions to achieve continuous heat-to-electricity conversion, yet the development of high-performance electrolytes is limited by the complex relationship between solvation of redox couple ions and thermopower. We present a data-driven machine learning strategy to identify high-performance solvents for thermogalvanic electrolytes. Our approach parametrizes the Seebeck coefficient by leveraging the strong correlation between DFT-derived ion-solvent interaction features and limited experimental data. A hybrid ML framework combining high-throughput DFT and interval regression was developed to screen extensive solvent candidates with SHAP analysis identifying polarity-related features as dominant factors. Experimental validation enabled model augmentation via a binary feature filter that incorporates solvation-relevant molecular features beyond two-body interactions. The framework validated known effective solvents and identified new candidates, which were confirmed via solvation structure calculations and UV-vis spectroscopy. This solvation-driven ML approach accelerates electrolyte discovery while providing fundamental insights into design principles for large solvation entropy differences.