Prediction Model of Converter Oxygen Consumption Based on Recursive Classification and Feature Selection

Oxygen consumption prediction for steelmaking converter is essential for optimal scheduling and energy saving of oxygen systems. To improve the prediction accuracy of oxygen consumption, an integrated prediction method based on feature space recursive division and feature selection is proposed. The feature space containing the whole converter production data is recursively divided into several feature subspaces containing the training subset. And the complexity of the data distribution will be reduced in each subspace. The simple data distribution will be more easily fitted by the prediction model. Based on recursive feature elimination, the appropriate feature variable combination and the corresponding oxygen consumption prediction models of the converter will be selected for each subset. For the test sample, it will be matched to a corresponding feature space by recursive division conditions. Then oxygen consumption is predicted by the corresponding prediction model based on the optimal combination of feature variables. A converter production data of a steel enterprise are used for testing. SVR and MLP will be used, respectively, for comparison in two groups of comparative experiments. The results show that the prediction performance of the integrated model is better than that of a single prediction model in multiple indicators.