MACHINE LEARNING APPLIED IN THE REDUCTION SIMULATION OF THE IRON ORE PELLETS COMPOSITE WITH RENEWABLE CARBON

The carbon insertion in iron pellets increases the reaction rate and decreases the reduction start temperature. This work aims to simulate reduction experiments of composite iron ore pellets with renewable carbon through Machine Learning. After a literature review, it was concluded that 28 features, ensembled into 4 classes, can influence the iron reduction percentual in the pellets. Decision Tree and Random Forest Regressors were applied to obtain a predictive model. The normalized data training was done (80% for training, 20% for test) using the K-Fold technique with 5 folds. The confidence interval of the predictions was calculated by the Bootstrap Resampling technique. Both models are satisfactory in the prediction because they shown adequate performance metrics (R² > 0.88 and MAE, MSE and RMSE of 0.001 order). However, the Random Forest Regressor is more convenient because it shown a narrower confidence interval for the predictions.The carbon insertion in iron pellets increases the reaction rate and decreases the reduction start temperature. This work aims to simulate reduction experiments of composite iron ore pellets with renewable carbon through Machine Learning. After a literature review, it was concluded that 28 features, ensembled into 4 classes, can influence the iron reduction percentual in the pellets. Decision Tree and Random Forest Regressors were applied to obtain a predictive model. The normalized data training was done (80% for training, 20% for test) using the K-Fold technique with 5 folds. The confidence interval of the predictions was calculated by the Bootstrap Resampling technique. Both models are satisfactory in the prediction because they shown adequate performance metrics (R² > 0.88 and MAE, MSE and RMSE of 0.001 order). However, the Random Forest Regressor is more convenient because it shown a narrower confidence interval for the predictions.