Dynamic feature selection for silicon content prediction in blast furnace using BOSVRRFE

Area of Science:

• Metallurgical Engineering
• Data Science
• Industrial Process Optimization

Background:

• Accurate silicon content prediction is crucial for blast furnace ironmaking efficiency.
• Industrial datasets present dynamic, complex, and nonlinear challenges for feature selection.
• Traditional static methods lack adaptability to changing operational conditions.

Purpose of the Study:

• To develop a dynamic feature selection algorithm for silicon content prediction.
• To enhance real-time optimization of input variables in industrial processes.
• To address limitations of traditional feature selection techniques in complex datasets.

Main Methods:

• Proposed a Bayesian online sequential update and support vector regression recursive feature elimination (BOSVRRFE) algorithm.
• Integrated Bayesian dynamic updating with recursive optimization for real-time feature importance adjustment.
• Validated the algorithm using industrial data from a large steel enterprise.

Main Results:

• The BOSVRRFE algorithm demonstrated superior performance in silicon content prediction compared to static methods.
• Achieved higher prediction accuracy, improved real-time adaptability, and enhanced model stability.
• Showcased rapid response to operational changes, supporting real-time industrial optimization.

Conclusions:

• BOSVRRFE offers an innovative approach to feature selection for complex, dynamic industrial data.
• Provides theoretical and practical guidance for improving silicon content prediction in ironmaking.
• Enables more effective real-time industrial prediction and process optimization.