Surrogate Models for CO₂ Utilization Pathways: Accelerating Industrial Park Design and Enterprise-Wide Optimization

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Sustainable Chemistry

Background:

• Carbon dioxide (CO2) utilization is crucial for mitigating climate change.
• Developing efficient computational tools for CO2 conversion processes is essential.
• Surrogate models can accelerate process simulation and optimization.

Purpose of the Study:

• To develop and validate surrogate models for methanol, ammonia, and urea production from CO2.
• To benchmark model performance across various data set sizes and statistical metrics.
• To assess the computational efficiency and predictive reliability of these surrogate models.

Main Methods:

• Utilized Aspen Plus simulations to generate data for surrogate model construction.
• Employed Latin Hypercube Sampling for data generation.
• Developed and evaluated linear and quadratic surrogate models using statistical metrics (R², adjusted R², predicted R², cross-validation R², RMSE, MAE) and confidence interval plots.

Main Results:

• Quadratic models better represent nonlinearities in methanol and ammonia production, especially with larger datasets.
• Linear models show strong generalization and efficiency for urea production.
• Predicted R² and cross-validation were critical in identifying overfitting in quadratic models at small sample sizes.
• Sufficient data sampling is vital for robust surrogate model performance.
• Surrogate models achieved significant prediction time savings compared to direct simulations.

Conclusions:

• Validated surrogate models provide a standardized benchmark for CO2 utilization processes.
• These models serve as reusable resources for research and can be integrated into industrial eco-park network models.
• The models reduce computational burden while maintaining predictive accuracy, supporting decarbonization initiatives.