Screening of Natural Oxygen Carriers for Chemical Looping Combustion Based on a Machine Learning Method

The screening of high-quality oxygen carriers is a key focus in the field of chemical looping combustion. However, the existing screening methods have the problems of being high cost and having long material design cycles. Here, a machine learning model has been established which successfully predicted the effect of composition, porosity, specific surface area, and other physicochemical properties on the redox performance. A database consisting of 190 samples was used to train the BP-ANN algorithm and the SVM algorithm. The SVM algorithm triumphs over the BP-ANN algorithm in that the best model by the SVM algorithm makes predictions with a high coefficient of determination (R 2 = 0.961) and a low root mean square error (RMSE = 0.014). According to the obtained model, the copper ore was estimated to exhibit high reaction performance in terms of 68% CH4 conversion and 96% CO conversion at 950 °C. We anticipate the machine learning method can be extended to predict the performance of oxygen carriers for other chemical looping applications.