96
Z. Liu et al.
oxygen consumption in converter steelmaking provides an important basis for the
optimal scheduling of the oxygen system and is of great significance to further save
energy and improve comprehensive economic benefits.
As the statistical learning method has been widely used in industrial intelligence, more and more scholars have studied the oxygen consumption prediction in
converter steelmaking based on a statistical learning method. Qin et al. [2] solve the
Support Vector Regression model (SVR) by particle swarm optimization algorithm,
which improves the prediction accuracy and generalization ability of the model;
Wang et al. [3, 4] combine the grey model with the BP network in parallel. They
weighted the sum of the results of individual models and improve the performance
of a single model; Zhao et al. [5, 6] optimize the extreme learning machine model
by improved genetic algorithm. And it improved the prediction performance of the
extreme learning machine; also, they combine the non-equidistant gray model with
the generalized regression neural network in series, and it showed some improvement effect; Li et al. [7, 8], based on oxygen decarbonization efficiency, used SVR
to predict oxygen consumption, respectively, in static and dynamic control stages.
And the oxygen consumption prediction performance of the model in two stages was
improved. Zhang et al. [9] embedded the grey model into the Elman neural network to
solve the problems of local optimization and over-fitting of ordinary neural networks.
To sum up, the statistical learning method has achieved some success in the oxygen
consumption prediction of the converter, and the hybrid model has become the development trend of energy prediction [10–12]. In the process of converter steelmaking,
the production data has a high dimension and strong correlation, which leads to
complex data distribution, so it is difficult for a single model to accurately fit the
whole data distribution. At present, in the research of converter oxygen consumption
prediction, the researchers mainly select the feature variables based on mechanism.
And it is difficult to select the optimal combination of feature variables for prediction.
For this problem, the complexity of data distribution is reduced by a recursive division
of feature space, and the optimal combination of feature variables will be selected
based on Recursive Features Eliminate (RFE). Firstly, based on the Decision Tree
Regression (DTR) idea, the feature space containing the whole data set is recursively
divided to feature subspace including data subset. Based on each feature subspace,
the corresponding optimal feature combination and prediction model based on the
statistical learning method are selected by the RFE algorithm. For the sample to be
predicted, according to the feature space division condition, it will be matched to
one of the feature subspaces. Then the corresponding model is used for prediction.
Converter Oxygen Consumption Analysis
In the process of converter steelmaking, oxygen is blown into the converter and redox
reaction occurs with the chemical elements (Fe, C, Si, P, Mn, and S, etc.) in the raw
materials of steelmaking (steel scrap and molten iron, etc.). This emits huge heat
and increases the temperature in the furnace. At the same time, it is necessary to
Z. Liu et al.
oxygen consumption in converter steelmaking provides an important basis for the
optimal scheduling of the oxygen system and is of great significance to further save
energy and improve comprehensive economic benefits.
As the statistical learning method has been widely used in industrial intelligence, more and more scholars have studied the oxygen consumption prediction in
converter steelmaking based on a statistical learning method. Qin et al. [2] solve the
Support Vector Regression model (SVR) by particle swarm optimization algorithm,
which improves the prediction accuracy and generalization ability of the model;
Wang et al. [3, 4] combine the grey model with the BP network in parallel. They
weighted the sum of the results of individual models and improve the performance
of a single model; Zhao et al. [5, 6] optimize the extreme learning machine model
by improved genetic algorithm. And it improved the prediction performance of the
extreme learning machine; also, they combine the non-equidistant gray model with
the generalized regression neural network in series, and it showed some improvement effect; Li et al. [7, 8], based on oxygen decarbonization efficiency, used SVR
to predict oxygen consumption, respectively, in static and dynamic control stages.
And the oxygen consumption prediction performance of the model in two stages was
improved. Zhang et al. [9] embedded the grey model into the Elman neural network to
solve the problems of local optimization and over-fitting of ordinary neural networks.
To sum up, the statistical learning method has achieved some success in the oxygen
consumption prediction of the converter, and the hybrid model has become the development trend of energy prediction [10–12]. In the process of converter steelmaking,
the production data has a high dimension and strong correlation, which leads to
complex data distribution, so it is difficult for a single model to accurately fit the
whole data distribution. At present, in the research of converter oxygen consumption
prediction, the researchers mainly select the feature variables based on mechanism.
And it is difficult to select the optimal combination of feature variables for prediction.
For this problem, the complexity of data distribution is reduced by a recursive division
of feature space, and the optimal combination of feature variables will be selected
based on Recursive Features Eliminate (RFE). Firstly, based on the Decision Tree
Regression (DTR) idea, the feature space containing the whole data set is recursively
divided to feature subspace including data subset. Based on each feature subspace,
the corresponding optimal feature combination and prediction model based on the
statistical learning method are selected by the RFE algorithm. For the sample to be
predicted, according to the feature space division condition, it will be matched to
one of the feature subspaces. Then the corresponding model is used for prediction.
Converter Oxygen Consumption Analysis
In the process of converter steelmaking, oxygen is blown into the converter and redox
reaction occurs with the chemical elements (Fe, C, Si, P, Mn, and S, etc.) in the raw
materials of steelmaking (steel scrap and molten iron, etc.). This emits huge heat
and increases the temperature in the furnace. At the same time, it is necessary to