Hierarchical Granular Computing-Based Model and Its Reinforcement Structural Learning for Construction of Long-Term...

Area of Science:

• Industrial Engineering
• Data Science
• Materials Science

Background:

• Byproduct gas is a critical energy source in the steel industry, with flow tendency crucial for production planning.
• Accurate numeric estimation and reliability through prediction intervals (PIs) are essential for practical applications.
• Long-term PIs provide valuable insights into future trends for operational management.

Purpose of the Study:

• To develop a hierarchical granular computing (HGrC)-based model for constructing long-term prediction intervals (PIs).
• To enhance the reliability and accuracy of byproduct gas flow predictions in the steel industry.
• To improve the efficiency and applicability of prediction models for both periodic and nonperiodic data.

Main Methods:

• A hierarchical granular computing (HGrC) model integrating probabilistic modeling and hierarchical information granularities.
• Monte-Carlo search with policy gradient for reinforcement structure learning to optimize model performance.
• Application of a parallel strategy to ensure efficiency for real-world industrial scenarios.

Main Results:

• The proposed HGrC model effectively constructs long-term PIs, offering numeric predictions with interval-valued reliability.
• Unequal granule sizes enable the model to handle both periodic and nonperiodic byproduct gas flow data.
• Experimental results show the method outperforms existing techniques in prediction accuracy and stability.

Conclusions:

• The HGrC model provides a superior approach for long-term PI construction in byproduct gas flow prediction.
• Reinforcement structure learning enhances model stability compared to other RL methods.
• The model's efficiency and adaptability make it suitable for practical steel industry applications.