Rational Design and Screening of Polyhydroxy Electrolyte Additives via Artificial Intelligence Framework for Ultrastable Zinc Anodes.

Electrolyte additives represent a cost-effective strategy to address Zn anode interfacial instability in aqueous zinc-ion batteries (AZIBs). However, conventional experimental trial-and-error is inherently inefficient, impeding commercialization progress. Herein, we propose an artificial intelligence-driven framework to rationally design and screen electrolyte additives. A graph neural network (GNN) is constructed to simultaneously predict multiple quantum chemical properties with high accuracy, eliminating costly calculations. Crucially, through systematic analysis of a series of cyclohexanol derivatives, we propose hydroxyl mass density (HMD) as a key structural descriptor. Unlike conventional approaches focusing solely on hydroxyl number while neglecting molecular weight, HMD captures the intrinsic density of functional groups, and we reveal its strong correlations with diverse molecular properties, elucidating the underlying theoretical mechanisms. Guided by HMD, inositol (INO), exhibiting the highest HMD and most balanced property profile, is identified as the optimal additive. INO synergistically reconstructs the Zn2 + solvation sheath, preferentially adsorbs on Zn facets, retards surface migration, lowers desolvation energy, and suppresses HER and corrosion. Consequently, the INO-modified electrolyte delivers ∼4000 h Zn||Zn cycling, 99.82% CE over 3200 cycles, and 1000-cycle full-cell stability. This work establishes a closed-loop paradigm from data-driven prediction to experimental validation, offering a transferable framework for rational electrolyte design.