(215c) Machine Learning of Carbon Electrodes for Electric Double Layer Capacitors

Electric double layer capacitors (EDLCs) (a.k.a. supercapacitors or ultracapacitors) have attracted great research interest over the past few decades because of their potential applications for electrical energy storage with high power density, long-term cycle stability, and product safety. However, the energy density of EDLC is typically smaller than that of an alternative energy-storage device such as Li-ion battery, pressing for further improvements by using better electrode materials and charge carriers. With extremely high porosity and specific surface area, porous carbon shows great promise to lift the upper limit of the energy densities of EDLCs, thereby expanding their practical usage. Despite a wealth of literature on experimental explorations, it is still unclear how the structure and microscopic features of carbon electrodes affect the EDLC performance. In this work, we use machine learning methods to establish quantitative correlations between the structural features of porous carbons and their performance (i.e., energy and power density) based on extensive experimental data from the literature. We demonstrate that machine learning is able to not only reproduce the structure-properties relationships of EDLCs at static conditions but also forecast the device performance far away from equilibrium (e.g., charging/discharging behavior at high-scan rate). Moreover, we have identified structural features of carbon electrodes that can expand the region of supercapacitor performance in the Ragone plot.