Neural Algorithm Aided Operation of CO₂ Electrolyzers

While the number of reports on the electrochemical carbon dioxide reduction increases at an ever-accelerating rate, achieving long-term stable, selective, and energy efficient operation is still challenging. This can be attributed mostly to the short length of lab-scale measurements and the complexity of cell operation parameters. Here we introduce a high-throughput cell operation testing methodology, including data evaluation and process optimization by machine learning algorithms. An autonomously operating test station allowed collection of enough data to develop an artificial neural network model. When the model is trained on a fraction of a large data set, predictions for the operation of the same cell under different conditions are very precise. Accurate predictions can also be made for newly assembled cells and at parameter settings outside of the training parameter space. Our results pave the way for the long-term stable operation of CO₂ electrolyzers by the adaptive optimization of the process conditions based on machine-learning-based holistic data evaluation.