Prediction of Specific Capacitance and Equivalent Series Resistance of Activated Carbon-Based Supercapacitors Using an Adaptive Neuro-Fuzzy Inference System

Supercapacitors are attracting growing interest as energy storage devices owing to their high power density and long cycle life. Their performance depends strongly on the electrode material, particularly activated carbon with its complex pore architecture. The relationship between characterization parameters and both specific capacitance (Cs) and equivalent series resistance (ESR) is non-linear and difficult to model conventionally. This study develops a predictive model based on an Adaptive Neuro-Fuzzy Inference System (ANFIS) using secondary data from forty samples reported in international scientific literature. Input variables comprise specific surface area (SBET), pore volume (Vp), pore diameter (Dp), crystallite size, diffraction angle, and surface functional groups, with the dataset partitioned at a ratio of 80:20 for training and testing. Evaluation demonstrates that ANFIS attains a coefficient of determination of 0.9851 for Cs and 0.9901 for ESR, surpassing Random Forest across every metric. ANFIS yields RMSE = 8.8 F/g and MAE = 4.5 F/g for Cs, and RMSE = 0.08 Ω and MAE = 0.04 Ω for ESR. Surface plot analysis reveals a synergistic interaction between SBET and Vp on Cs and the effect of crystal structure on ESR. The model captures complex relationships accurately while remaining interpretable through membership functions and fuzzy rules.