Bayesian Symbolic Learning to Build Analytical Correlations from Rigorous Process Simulations: Application to CO₂...

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Data Science

Background:

• First-principles process models are essential but often complex, costly, and difficult to use due to convergence issues.
• Conventional surrogate models lack the interpretability and algebraic manipulability of analytical expressions.

Purpose of the Study:

• To demonstrate the application of Bayesian symbolic learning for deriving simplified, closed-form process models.
• To enhance the accessibility and utility of process models for experimental research and development.

Main Methods:

• Utilized Bayesian symbolic learning to analyze synthetic data from rigorous process simulations.
• Applied the method to CO2 capture processes simulated using Aspen HYSYS.
• Derived simplified, interpretable equations for key economic and environmental performance variables.

Main Results:

• Successfully generated accurate, closed-form analytical expressions from complex process simulations.
• The derived equations offer improved interpretability and algebraic tractability compared to traditional surrogate models.
• Identified key variables influencing the economic and environmental performance of CO2 capture processes.

Conclusions:

• Bayesian symbolic learning offers a powerful approach to streamline process modeling, making it more accessible.
• The derived analytical expressions facilitate deeper process insights and comparative analysis.
• This method enables effective benchmarking of emerging technologies against established ones, like CO2 capture.