Data-driven discovery of carbonyl organic electrode molecules: machine learning and experiment

Abstract

Carbonyl organic electrode molecules have broad prospects for application in lithium-ion batteries due to their environmental friendliness and cost-effective merit. To overcome the drawbacks associated with traditional time-consuming and costly trial and error experiments, herein, high-throughput calculations and machine learning methods have been employed to accelerate the development of high-performance carbonyl organic electrode molecules by evaluating one million molecules. Hierarchical clustering has been introduced into the selection process to find those target molecules and help us eliminate non-ring molecules. As the reduction potential is a crucial factor in evaluating the performance of electrode materials, based on the created dataset of organic electrode molecules by high-throughput calculations, we have built a machine learning model whose coefficient of determination can reach 0.88 for predicting the reduction potential. With the above efforts, naphthalene-1,4,5,8-tetraone with high reduction potential and energy density has been screened out and indeed exhibits a long cycle life of 2500 cycles at 1 A g⁻¹ and a high discharge voltage of 2.5 V. The approach developed in this work offers new insight to filter advanced organic electrode molecules accurately and rapidly for Li-ion batteries.