Abstract: Developing high-performance and low-cost oxygen carrying materials is the key to the future commercial application of chemical-looping processes. Over 1 000 materials have been tested as the oxygen carriers for chemical-looping processes. Among them,ores and industrial by-products as oxygen carriers have recently attracted much attention recently due to their low cost and availability,especially the sufficient reactivity with the solid fuels. However,these materials have highly variable compositions,which strongly influences the performance in chemical-looping. It costs a lot to test one by one. Taking natural manganese ore as the object,and taking the existing experimental data as the training set,the new machine learning algorithm was used to predict the performance of manganese bearing minerals in chemical chain reaction,and compare the performance of support vector machine and artificial neural network in the prediction process.The results show that the two algorithms have good accuracy for the data in the training set,but there are differences in the accuracy and robustness of the new input value prediction. Support vector machine can avoid the over fitting phenomenon of neural network in small sample training set. The sensitivity analysis shows that the change of manganese content in oxygen carrier has a great influence on the reaction characteristics,while the effect of specific surface area is small.