Machine learning aided cyclic stability prediction for supercapacitors

Abstract

The rapid depletion of fossil fuel resources and vast demands for electricity has prompted scientists to think of increased reliance on renewable energy sources and, therefore, next-generation energy storage devices (including supercapacitors). A supercapacitor's performance depends on its intrinsic features such as electrode materials, type of electrolytes, etc. This paper uses machine learning algorithms to find a correlation between these inherent features and supercapacitor performances in terms of cyclic stability. A considerable amount of data on supercapacitor cyclic stability and other relevant features from 400+ published papers are collected, and different ML algorithms have been used to build models. Attribute prioritization and Principal Component Analysis (PCA) are performed to remove redundant features, reduce the computation time, and increase the interpretability of the dataset. Key material insights based on this study are (i) Ni and Co-based materials are among the most studied materials, (ii) electrodeposited Ni(OH)2 on Ni foam is a combination that is industrially relevant, exhibits high initial specific capacitance but has poor capacity retention, and (iii) 3 dimensional nanoparticles are preferable to lower dimensions as electrode materials for cyclability. These insights, among others given here, offer ways forward to pursue new materials combinations for achieving high cyclic stability.