A transfer predictive control method based on inter-domain mapping learning with application to industrial roasting...

Area of Science:

• Process Control
• Machine Learning
• Chemical Engineering

Background:

• Roasting temperature is critical for process operations.
• Model predictive control (MPC) is used for temperature stabilization.
• Feed composition changes cause data distribution shifts, leading to model mismatch in MPC.

Purpose of the Study:

• To propose a novel transfer predictive control method for roasting temperature control.
• To address the challenge of model mismatch caused by data distribution differences.
• To enhance the accuracy and robustness of online process control.

Main Methods:

• Developed an inter-domain mapping learning-model predictive control (IDML-MPC) approach.
• Treated historical and online data as distinct domains for mapping.
• Learned a distribution mapping function to align historical data with online data distributions.
• Built an accurate online prediction model for control by minimizing a cost function.

Main Results:

• The IDML-MPC method demonstrated effectiveness in comparative experiments.
• Experimental results showed significant improvement over state-of-the-art methods.
• The proposed method achieved an average 56.98% reduction in root mean square error for roasting temperature control.

Conclusions:

• Accounting for distribution differences between historical and online data is crucial for changing production conditions.
• The IDML-MPC method offers a robust solution for roasting temperature control under varying operational circumstances.
• This approach enhances control performance and reduces prediction errors in industrial processes.