Machine learning-assisted prediction of water adsorption isotherms and cooling performance

Abstract

Water adsorption in porous adsorbents has drawn considerable attention for its tremendous potential in numerous environment- and energy-related applications. However, owing to the huge experiment or computational cost, it is still an extremely challenging task to rapidly obtain the water adsorption isotherms of a large number of adsorbents. In this work, machine learning (ML) models for water adsorption isotherm estimation were developed based on the collected data from experimentally measured water adsorption isotherms of various adsorbents. It is demonstrated that the water adsorption isotherms can be successfully predicted by the random forest (RF) model, based on which the performance of water adsorption-driven applications such as adsorption cooling, water harvesting and water desalination can be quickly obtained. Taking adsorption cooling as an application example, an ML-based model based on extracted descriptors from predicted isotherms was developed to achieve the high-accuracy prediction of the specific cooling effects (SCE) and coefficient of cooling performance (COP_C) of a large number of adsorbent/water working pairs, based on which the relationship between structural properties of adsorbents and cooling performance was also extracted. This work opens up the possibility of the use of ML to efficiently predict water adsorption isotherms of numerous adsorbents, which may not only accelerate the discovery of potential adsorbents for water adsorption but also the development of high-performing water adsorption-driven systems.