Hot Rolling Line Material Prediction by Classified Regression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hot rolling line control systems face high computational load and decreased prediction accuracy when predicting materials for multiple steel grades, as they rely on collecting and using data close to the input values, leading to inefficient data extraction and regression models that struggle with data far from the collected range.
Innovation Solution
A hot rolling line control system that includes a classification criteria creation and material model regression unit, which classifies and regresses operation and material measurement data to create a material model for each classification, using a classification-property learning approach to accurately predict materials while managing calculation load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data close to the input value is collected for material prediction, then prediction accuracy for close data is improved, but prediction accuracy for data far from the collected range decreases
Solution Approach 1:
The patent segments the data space by dividing it into multiple regions based on distance from reference points. Different prediction strategies are applied to different regions: data close to reference points uses direct reference, while data far from reference points uses extrapolation or alternative reference points. This segmentation allows the system to optimize for both close and far data prediction accuracy.
Solution Approach 2:
The patent changes the parameter of data selection from purely distance-based to a hybrid approach that considers both distance and data distribution characteristics. By adjusting the weighting between close data and far data, and by dynamically selecting reference points based on the input characteristics, the system adapts to different prediction scenarios and maintains accuracy across the full range.
2Measurement precision
If data extraction processing is performed for each material prediction point, then prediction accuracy is maintained, but computer load becomes high
Solution Approach 1:
The patent performs preliminary actions by pre-processing and organizing the reference data into an efficient data structure before actual prediction. Reference points are pre-selected and stored with their associated data, creating a ready-to-use reference library. This preliminary organization significantly reduces the computational load during actual prediction operations while maintaining accuracy.
Solution Approach 2:
The patent extracts only the essential and most relevant features from the raw data to create a compact representation. By extracting key characteristics and storing only necessary information in the reference data structure, the system reduces memory usage and processing requirements while preserving the information needed for accurate prediction.
3Measurement precision
If only close data is collected for regression, then local prediction accuracy is improved, but model generalization capability decreases
Solution Approach 1:
The patent implements a dynamic data selection mechanism that adapts the composition of training data based on the specific prediction task and input characteristics. Rather than statically using only close data, the system dynamically adjusts which data points to include, balancing between close data for local accuracy and far data for generalization, thereby improving both local precision and model robustness.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A hot rolling line control system includes a rolling condition setting unit 21 that sets rolling conditions, an operation data collection unit 22 that collects rolling conditions and operation data of a line during rolling, an operation data storage unit 23 that stores operation data, a material measurement data storage unit 24 that stores material actual measurement data obtained by measuring a material of a rolled steel sheet 13, a material prediction unit 26 that predicts material of rolled steel sheet 13, and a material prediction data storage unit 27 that stores material prediction data in the material prediction unit 26, and the material prediction unit 26 includes a classification criteria creation and material model regression unit 260 that creates classification criteria using operation data and the material actual measurement data, classifies the operation data and the material actual measurement data according to the created classification criteria, and regresses the classified operation data and material actual measurement data to create a material model for each classification.