Deep learning driven ionic liquid screening for toluene and styrene capture

Abstract

As typical aromatic volatile organic compounds (VOCs), toluene and styrene present significant health and environmental risks. Although ionic liquids (ILs) have emerged as promising absorbents for aromatic VOCs capture, the myriad diversities of ILs and limited experimental data availability pose critical challenges in efficiently identifying optimal candidates. In this work, advanced deep learning (DL) models are developed to predict the infinite dilution activity coefficients (γ^∞) of IL-toluene/styrene systems, which could act as the thermodynamic estimator of absorbent performance. By leveraging the transfer learning strategy, the models are first pre-trained on experimental data from similar system or computational data, then fine-tuned using experimental data points to improve predictive performance. A multi-level screening framework that considers physical and thermodynamic properties, as well as toxicity, is employed to identify favorable ILs from 170 commercially available candidates. Experimental validation and mechanism analysis are conducted to corroborate the excellent absorption performance of screened ILs. This research provides valuable guidance for artificial intelligence-aided design of absorbents towards other VOCs capture.