Optimization and predictive performance of hybrid carbon nanotubes-graphene electrodes- based supercapacitor for electric vehicles and electronics devices

The paper proposed an excellent machine learning based approach for designing and optimizing hybrid graphene-carbon nanotube electrodes for supercapacitors. In addition to utilizing recursive feature elimination with cross-validation (RFECV), which decreased the number of input variables from 40 to 12 and increased the accuracy of prediction, the authors also utilized non-dominated sorting genetic algorithm-II (NSGA II) to achieve an increase of 25% in energy density while achieving a high-power density greater than 10 kW/kg. Also, the authors developed new configurations of material through the use of variational autoencoders (VAE) that resulted in a 20% improvement in the specific capacitance of these configurations. By including gradient boosted regression trees (GBRT) into this research, the authors were able to include real world synthesis variable into synthesis process, increasing the accuracy of prediction (from RMSE = 0.12 to RMSE = 0.10). Additionally, by using long short-term memory networks (LSTM) the authors were able to accurately predict cyclic stability trends, with a correlation of 95% to experimental data. Finally, the authors used SHapley Additive exPlanations (SHAP) and Monte Carlo dropout to provide both interpretability and reliability. Overall, this work provides a robust predictive and generative framework that will advance the hybrid supercapacitor electrode design and accelerate its use in future generation of energy storage systems.