A Multiobjective Evolutionary Nonlinear Ensemble Learning With Evolutionary Feature Selection for Silicon Prediction...

Area of Science:

• Metallurgical Engineering
• Machine Learning
• Artificial Intelligence

Background:

• Accurate prediction of silicon content in molten iron is crucial for blast furnace ironmaking efficiency.
• Existing methods often fail to capture complex nonlinear feature relationships.

Purpose of the Study:

• To propose a novel multiobjective evolutionary nonlinear ensemble learning model with evolutionary feature selection (MOENE-EFS).
• To enhance the accuracy and stability of silicon content prediction in molten iron.

Main Methods:

• Developed a multiobjective evolutionary nonlinear ensemble learning model (MOENE-EFS) using extreme learning machines as base learners.
• Implemented a modified nondominated sorting differential evolution algorithm for optimizing accuracy and diversity.
• Employed a nonlinear ensemble method for combining base learners.

Main Results:

• The proposed MOENE-EFS model significantly improves prediction accuracy and stability.
• Evolutionary feature selection effectively identifies essential features, aligning with expert knowledge.
• MOENE-EFS outperforms existing models on benchmark and industrial data.

Conclusions:

• The MOENE-EFS model offers a promising approach for silicon content prediction in blast furnace ironmaking.
• Evolutionary feature selection and nonlinear ensemble methods are key to enhanced prediction performance.
• This approach addresses limitations of linear methods in handling complex industrial data.