Data-driven discovery of low temperature denitration catalysts with strong anti-sulfur and anti-water property

The data-driven discovery of low temperature denitration catalysts is studied in the present work with active machine learning (ML). Four acquisition functions of Bayesian global optimization are used to screen candidate catalysts predicted by the Gaussian process regression (GPR) model, and each acquisition function recommends one candidate for experiment. The new experimental results are fed back into the dataset to update the model. The Mn2Ce2V7/TiO2 catalyst is discovered after nine iterations of active leaning and the catalyst exhibits a denitrification efficiency of 81.88% in the testing gas mixture containing 1000 ppm NOx, 280 ppm SO2 and 6% H2O at 160 °C.