A feature-guided closed-loop optimization framework via stepwise PSO for process simulation: A case study on membrane-assisted methanol synthesis

초록

Optimization of complex chemical processes under data-scarce conditions remains highly challenging due to strong nonlinearity, multivariable coupling, and hierarchical parameter interactions. Conventional surrogate-assisted optimization frameworks typically rely on static models and one-shot global search strategies, which often result in limited robustness, inefficient exploration, and sensitivity to initial sampling. To address these limitations, this study proposes a feature-guided closed-loop optimization framework, termed the Stepwise PSO-based Package for Optimization of Targeted process simulation (SPOT). The framework systematically integrates machine learning-based surrogate modeling, feature importance diagnostics, uncertainty-aware objective formulation, and staged optimization within an adaptive closed-loop structure. In particular, a stepwise PSO algorithm is developed to sequentially optimize dominant and secondary decision variables based on the identified feature hierarchy, thereby improving search efficiency and mitigating premature convergence. The framework iteratively updates surrogate models and optimization trajectories through data feedback, enabling data-efficient exploration of complex design spaces. Its effectiveness is demonstrated through a case study on membrane-assisted methanol synthesis via CO2 hydrogenation. Compared with conventional PSO, the proposed framework exhibits enhanced convergence stability, reduced sensitivity to initial conditions, and improved optimization performance, achieving simultaneous improvements in CO2 conversion and exergy efficiency. Owing to its process-agnostic and adaptive design, the SPOT framework provides a generalizable and data-efficient methodology for optimization of complex chemical and energy systems. © 2026 The Authors

키워드