Data-driven adaptive modeling method for industrial processes and its application in flotation reagent control.

Area of Science:

• Process Control
• Data-Driven Modeling
• Machine Learning

Background:

• Industrial processes face evolving 'excitation' patterns due to input disturbances, leading to model mismatch.
• Existing adaptive process models struggle to incrementally learn new patterns without degrading performance on old data.
• Data-driven models require adaptation to maintain accuracy in dynamic industrial environments.

Purpose of the Study:

• To develop a data-driven process model with incremental learning ability for adapting to new process patterns.
• To enable adaptive process models to learn new patterns without performance degradation on historical data.
• To implement and validate an adaptive neural network process model for flotation reagent control.

Main Methods:

• A novel incremental learning method was proposed for updating data-driven process models.
• An adaptive neural network process model was developed utilizing the proposed incremental learning method.
• A nonlinear model predictive control strategy was implemented using the adaptive process model for flotation reagent control.

Main Results:

• The developed adaptive process model successfully accommodates new process 'excitation' patterns.
• The model preserves its performance on previously learned patterns while adapting to new ones.
• Industrial experiments in a lead-zinc froth flotation plant demonstrated the controller's practical viability.

Conclusions:

• The novel incremental learning method enables adaptive process models to handle evolving industrial dynamics.
• The adaptive neural network process model offers a promising solution for robust process control.
• The developed controller shows significant potential for practical application in flotation reagent control systems.