“Crypton 1.0”: Accurate cyclic voltammetry forecasting of activated carbon electrode with machine learning

Abstract

Cyclic voltammetry (CV) is a technique for determining the electrochemical properties of the electrode, and electrolyte in electrochemical systems. However, it is sensitive to various feature, and the correlation between them is not fully explained hitherto. An artificial neural network (ANN) was employed to create a CVs prediction for further explain the electrochemical properties in “water-in-salt” electrolyte e.g., scan rate, electrolyte concentration, and potential window. The electrochemical assistant software based on the developed model are then present (namely “Crypton 1.0″ beta version). The designed network architecture consists of two hidden layers with fixed number of neurons in the latter to reduce the calculation burden for loop training. Five-fold cross validation and single loop training with a variation of hidden neurons in the first hidden layer from 1 to 20 neurons were applied to generalize the prediction. The training was performed in batches corresponding to positive scan, negative scan, and full cycle scan to achieve comprehensive models. The final predictions are the product of averaged models with coefficient of determination (R2) over 0.98 for each scanning characteristic. Interestingly, the prediction for wide potential windows showed superior accuracy comparable with the CV from experimental measurement. In addition, the electrochemical stability window has been investigated, and found to be increased along with the electrolyte concentration explaining the concepts of using “water-in-salt” electrolyte. This work covers the understanding of electrochemistry and the model development via ANN for software development. Therefore, our work could be an alternative approach to reduce the experimental burden in the future development of electrochemical applications.