A deep learning protocol for analyzing and predicting ionic conductivity of anion exchange membranes.

Possessing high ionic conductivity is required to polymer-based membrane electrolytes. However, it is a challenge to evaluate the conductivity based on the structure of the polymer membrane without any measurements. We present a deep learning protocol to predict the hydroxide ion (OH-) conductivity from chemical structure information of poly (2,6-dimethyl phenylene oxide)-based anion exchange membranes (AEMs) grafting with one kind of functional cationic group. The modeling process includes data collection and feature processing, functional cationic group identification, OH- conductivity prediction and scientific law extraction. The established model achieves 99.7% of accuracy for classifying various functional cationic groups. The prediction error in OH- conductivity is ± 0.016 S/cm for quaternary ammonium based AEMs, ± 0.014 S/cm for saturated heterocyclic ammonium based ones, and ± 0.07 S/cm for those possessing imidazolium cations. The proposed protocol is powerful to assist researchers in designing the AEMs with predictable OH- conductivity, and provides a new research paradigm of the AEMs preparation. [Display omitted] • Construction of a deep neutral network model by using an experimental dataset. • Identify functional cationic groups and predict conductivity of anion exchange membranes. • Superior discernibility and predictive power on external dataset. • Feasibly extract scientific rules of specific polymer based membrane electrolytes. [ABSTRACT FROM AUTHOR]