The lack of clear guidance for designing novel aqueous electrolytes with wide electrochemical stability window (ESW) and strong resistance to hydrogen evolution reactions (HER) has hindered the development of safe and energetic aqueous batteries (ABs). Despite advancing scientific discovery by uncovering complex patterns, machine learning remains challenging for electrolyte design, owing to intricate additive formulations, solvation interactions, and coupled performance metrics, demanding chemically interpretable workflows. Herein, a multilevel artificial intelligent (AI) framework is developed for accelerated electrolyte design for ABs. The multi-task neural network is first applied to establish the elemental features of aqueous electrolytes with wide ESW, followed by a classification-regression model to identify additives with strong HER inhibition effect. Unsupervised learning combined with molecular dynamics simulations further provides a chemical explanation. Particularly, the elaborated additives with large polar topological structures reduce HER activity and expand ESW by enhancing the water confinement effect. As a proof of concept, experimental analyses further validate the long lifespan, evidenced by symmetric-cell cycling beyond 1100 h and more than 2500 cycles in Zn||VO2 full cells, along with the reliable operation of a 1.66 Ah punch-type device. This multilevel AI framework integrated with experimental validations should accelerate and rationalize the development of ABs.
Li et al. (2026) studied this question.