Risk-sensitive collaborative parameter tuning via calibrated deep surrogates for rare-earth electrolysis energy...

Area of Science:

• Chemical Engineering
• Process Optimization
• Data Science

Background:

• Rare-earth molten salt electrolysis faces challenges in energy efficiency due to process nonlinearity and operational constraints.
• Existing data-driven methods often ignore predictive uncertainty, leading to risks of constraint violations.

Purpose of the Study:

• To develop a risk-sensitive framework for collaborative parameter tuning in rare-earth electrolysis.
• To enhance energy efficiency while ensuring operational safety and minimizing constraint violations.

Main Methods:

• Utilizing calibrated deep surrogate modeling for predicting energy efficiency and quantifying uncertainty.
• Employing uncertainty-aware constrained Bayesian optimization for risk-sensitive parameter tuning.
• Implementing a deep ensemble surrogate with a calibration procedure for reliable uncertainty estimation.

Main Results:

• The proposed framework demonstrated faster convergence compared to uncertainty-agnostic methods.
• Achieved higher energy-efficiency improvements in rare-earth electrolysis.
• Significantly reduced the probability of constraint violations during operation.

Conclusions:

• The risk-sensitive framework effectively balances energy efficiency and operational safety in rare-earth electrolysis.
• Calibrated deep surrogate modeling and uncertainty-aware optimization are crucial for reliable process tuning.
• The approach offers a robust solution for complex industrial processes with strict constraints.