(733b) Predicting Reaction Mechanism of Electrochemical CO2 Reduction Using Machine Learning

Understanding the mechanism of an electrochemical reaction could, in principle, be one of the important factors in the accelerated discovery of highly efficient electrocatalysts. Cyclic voltammetry (CV) is one of the techniques providing important mechanistic insight into an electrochemical reaction by the virtue of the change in current density when a reactant or an intermediate species is adsorbed/desorbed on the electrode surface. However, interpreting a CV can be influenced by subjectivity from the experience of the electrochemist. Moreover, quantifiable confidence in such interpretations is wanting. Machine Learning (ML) algorithms can be applied in such cases to circumvent the heuristics in CV interpretations and user subjectivity to present robust, quantifiable confidence in predicting electrochemical reaction mechanisms. One of the most attractive electrochemical reactions in recent years has been the electrochemical CO₂ reduction (CO2R) to value-added chemicals due to its potentially high impact on the industries, society, and the environment. This work shows the implementation of an ML model (Neural networks) to predict the reaction mechanism of CO2R by learning from different data sets of cyclic voltammograms. The simulated and experimental cyclic voltammograms are used to train the neural network to identify the dependence on the parameters like CV scan rates and uncompensated resistance. For each individual CV, the voltammogram serves as an input to the model. The expected output predicts the likelihood of the occurrence of either simple reversible electron transfer reactions or irreversible product formation after an electron transfer reaction. This ML model confirms that the predictions can be made even by simple models as a substitute for the conventional subjective approach.