Multivariable Collaborative Modeling With Knowledge Transfer and Its Application in Soft Sensing of Iron Flotation...

Area of Science:

• Mineral Processing
• Data Science in Industrial Operations
• Machine Learning for Process Optimization

Background:

• Iron flotation production stages frequently change due to equipment and raw material updates.
• Existing operating condition prediction models become obsolete with stage changes, wasting valuable historical data.
• Data-driven models built on small current datasets lack accuracy.

Purpose of the Study:

• To propose a knowledge transfer method for leveraging historical data in updated production stages.
• To enable rapid establishment of accurate prediction models with minimal new data collection.
• To enhance tailings grade soft sensor accuracy by incorporating production process parameters.

Main Methods:

• A knowledge transfer approach is utilized to adapt existing models to new production stages.
• A multivariate collaborative modeling strategy is introduced for soft sensor development.
• The method integrates historical datasets with limited current-stage data for model building.

Main Results:

• The proposed knowledge transfer method effectively utilizes outdated datasets for new operating conditions.
• Accurate prediction models are established rapidly, reducing the need for extensive new data collection.
• The multivariate approach significantly improves the accuracy of tailings grade soft sensors.

Conclusions:

• Knowledge transfer is a viable strategy for updating prediction models in dynamic industrial processes like iron flotation.
• Integrating production process parameters into soft sensor models enhances their predictive capabilities.
• The method demonstrates practical effectiveness through experimental validation and industrial application.