Gray-box surrogate models for flash, distillation and compression units of chemical processes

Surrogate models of chemical processes can substitute rigorous models that are computationally expensive or of limited stability by simplified and typically solely data-driven models. In this work, gray-box surrogate models of classical process engineering unit operations comprising flash, distillation and compression units are developed to provide accurate models that allow for fast and stable predictions in view of later optimization of coupled models. The gray-box surrogates are first tested as individual models and then applied to model the cracked-gas compression of an ethylene plant, including a recycle stream. The process streams are hydrocarbon mixtures containing 50 components, which typically leads to significant convergence issues with rigorous approaches. A concluding comparison of the proposed surrogate models’ accuracies proves their robustness and computational efficiency and highlights the advantages of the proposed modeling methodology that complements and extends simple but physically meaningful white-box models with black-box models from the field of machine learning.