Construction of molecular model-driven hybrid model for atmospheric distillation separation of crude oil and dynamic prediction study of product composition distribution

Distillation is crucial for crude oil separation but poses modeling challenges due to thermodynamic complexities and strong nonlinearities in multicomponent systems. This study proposes a dynamic hybrid modeling approach to predict the time-dependent molecular composition distribution of separated products in atmospheric crude oil distillation processes. A benchmark model system incorporating 152 real molecular species is developed. By integrating a dynamic first-principles model for the Pre-Flash column, first-principles models for the atmospheric column’s top and bottom sections, and a surrogate model for the main column, a dynamic hybrid simulation framework for atmospheric distillation is established. Validation under steady-state conditions demonstrates strong agreement between the hybrid model and Aspen Plus simulations, with absolute errors for all product components remaining below ± 5 %. The model achieves high predictive accuracy for molecular mole fractions and vapor–liquid equilibrium constants. Under dynamic operating conditions, the model demonstrates robust stability, with product component errors consistently maintained within ± 5 % upon re-establishing steady state. Over 90 % of molecular content predictions show absolute errors between −0.5 % and 1.0 %. Predicted vapor–liquid equilibrium constants align closely with reference values, as evidenced by data points distributed uniformly along the parity line. Experimental results confirm the hybrid model’s capability to capture dynamic characteristics of crude oil distillation, demonstrating both high accuracy in molecular-level product distribution predictions and practical engineering applicability.