Machine learning models for capacitance prediction of porous carbon-based supercapacitor electrodes

Abstract

Energy storage devices (ESD) and systems (ESS) with better performance, higher reliability, longer life, and wiser management strategies are needed for daily technology advancement. Among these devices, the supercapacitor is the most preferable due to its high-limit capacitance that esteems more than different capacitors. Today, it is considered a significant challenge to design high-performance materials for supercapacitors by exploring the interaction between characteristics and structural features of materials. Advanced control strategies depend on the status of numerous indicators, but creating such systems requires making trade-offs between a vast array of elements. From an engineering perspective, Specific capacity is regarded as the most crucial performance-related characteristic of supercapacitor electrodes. Because of this, it is essential to predict capacitance when assessing a material's potential for use in constructing supercapacitor-electrode applications. Machine learning (ML) can significantly speed up computation, capture complex mechanisms to enhance the accuracy of the prediction and make the best choices based on detailed status data. We aimed to develop a new strategy for the assisted design of high-performance supercapacitor materials by applying ML to analyze the relationship between capacitance and structural features of porous carbon materials (PCMs) using hundreds of experimental data in the literature. To predict the associated capacitance performance for the porous carbon electrode, structural characteristics of the electrode material, Several elements of the electrochemical test, including specific surface area, Potential Window, pore volume, and Electrolyte material, as well as data extracted from nearly 100 practical research papers that were published, were used. In the present study, Linear Regression (LR), Regression Tree (RT), and Adaptive Neuro-Fuzzy Inference System (ANFIS) were used to estimate the SC’s capacitance. The effectiveness of the ML models was evaluated in terms of the root mean square error (RMSE), mean absolute error (MAE), and the correlation between expected yield and system-provided yield. The developed ANFIS model, with RMSE, MAE,and R values of 22.8, 39.7647, and 0.90004, respectively, compares favourably regarding prediction performance to other models built for this purpose.