Applying machine learning to understand the properties of biomass carbon materials in supercapacitors

Abstract

Carbon is a fundamental material in developing electrochemical double-layer capacitors (EDLCs), also known as supercapacitors. Many studies have proven the impact of various carbon material properties, such as surface area, pore volume, and chemical surface composition, on the specific capacitance of supercapacitors (EDLCs). However, research endeavors to comprehensively evaluate the contribution of these material properties in correlation with experimental parameters, such as electrolyte concentration, voltage window, and current density, are scarce. This study aimed to employ machine learning algorithms to comprehend the interdependence between the properties of biomass-based carbon and the aforementioned experimental parameters with the capacitance of EDLCs. Four models of the machine learning algorithms were utilized in this study, including linear regression (LR), M5-Rules, Random Tree (RT), and Multi-Layer Perceptron (MLP), to determine the most suitable algorithm for analyzing and predicting the capacitance of EDLCs. The results revealed that the MLP model exhibited the highest determination correlation coefficient (R) of 0.871 with a Mean Absolute Error (MAE) of 45.069 F/g. Besides, the study utilized a machine learning correlation attribute model and observed that the supercapacitor’s surface area and pore volume demonstrated significant correlations with the same correlation ratio of 0.4. In conclusion, these findings emphasize the importance of considering surface area and pore volume in developing and optimizing supercapacitors. Finally, this study adds knowledge in supercapacitors and provides valuable insights for designing and developing high-performance energy storage devices.