Prediction of energy storage capability of carbide-derived carbon materials using non-linear Mt-QnSPR approach

Nanoporous carbon-based materials are being investigated as potential electrical double-layer-based ultracapacitors. The electrochemical properties of nanoporous carbon materials strongly depend upon their structure. Carbide-derived carbon (CDC) materials are considered promising carbon-based energy storage because of their diverse structural variations with high content of micropores, surface area, and pore size distribution. Considering all these facts, therefore, for the first time, a multi-target quantitative nanostructure property relationship (Mt-QnSPR) approach was used on a set of carbide-derived carbon materials to predict the electrical double-layer capacitance in non-aqueous electrolyte. Here, the volumetric capacitance of these carbon electrodes was predicted in terms of cathodic capacitance (Cv_NEG) and anodic capacitance (Cv_POS) values using a single multi-target ANN model. Two models, one with experimentally derived structure descriptors and the other with descriptors derived from the Monte Carlo method, were developed and tested using an external test set. The prediction abilities of both models were compared using various statistical parameters. The results showed that both models were quite robust and reliable (R²_test = 0.962, ΔRm² = 0.038. CCC = 0.977, IIC = 0.872, RMSE = 2.610 for 5–5-2 ANN model; R²_test = 0.858, ΔRm² = 0.044, CCC = 0.925, IIC = 0.708, RMSE = 4.77 for 2–2-2 ANN model). But 5–5–2 model outperforms 2–2–2 model in terms of training and prediction ability. The applicability domain of these models was also verified using the leverage approach.