Effects of Surrogate Hybridization and Adaptive Sampling for Simulation-Based Optimization.

Area of Science:

• Chemical Engineering
• Computational Science
• Optimization

Background:

• Process simulators are crucial for complex modeling but optimization is hindered by high costs, lack of equations, and convergence issues.
• Surrogate modeling and surrogate-based optimization offer solutions, with black-box and hybrid surrogates being common approaches.

Purpose of the Study:

• To assess and compare two main optimization methodologies: fixed a priori sampling with deterministic solvers and adaptive sampling-based optimization.
• To systematically evaluate the impact of black-box versus hybrid surrogates on optimization performance.
• To analyze the influence of sampling quantity, dimensionality, formulation, and hybridization on solution convergence, reliability, and CPU efficiency.

Main Methods:

• Comparison of optimization using surrogates trained on fixed samples versus adaptive sampling strategies.
• Systematic evaluation of black-box surrogates against hybrid surrogates employing a model-correction architecture.
• Testing across mathematical benchmarks (up to ten dimensions) and engineering case studies (extractive distillation, adsorption).

Main Results:

• Hybrid modeling enhances surrogate robustness and reduces solution variability, albeit with increased optimization costs.
• Adaptive sampling methods demonstrate superior efficiency and consistency compared to fixed-sampling strategies.
• The study quantifies the effects of sampling, dimensionality, formulation, and hybridization on optimization outcomes.

Conclusions:

• Hybrid surrogates offer improved reliability and reduced variability in process simulation optimization.
• Adaptive sampling is a more efficient and consistent approach for surrogate-based optimization than fixed-sampling methods.
• The findings provide valuable insights for optimizing expensive and complex process simulations effectively.