Objective Batch pipelining of product oils inevitably leads to oil mixing at the junction of oils transmitted in sequential batches within the pipelines. This disruption affects the formulation of station dispatching plans and offtake schemes. Therefore, accurately predicting the concentration distribution of oil mixing sections is considered a crucial foundation for enhancing the control of oil mixing, reducing energy consumption for mixed oil treatment, and preventing oil quality incidents at stations. Despite this, multi-dimensional numerical models of oil mixing have exhibited several drawbacks. These include lengthy calculation processes, inefficiencies in long-distance pipeline applications, neglect of the oil mixing progression mechanism process due to traditional data-driven approaches, and violations of physical principles, low accuracy, and poor interpretability in their results.

Methods Through analyzing the oil mixing progression mechanism, this study delved into the fundamental control equations and initial boundary conditions relevant to oil mixing progression. A physics-guided coupling loss function was constructed by coupling the automatic differential method with a deep learning model, enabling the confinement of model prediction results within the corresponding physical solution space of oil mixing progression. Following this, the initial numerical solutions for short pipeline sections were used to predict the mixed oil concentration distribution throughout long-distance transmission pipelines based on the recursive rolling prediction and data correction methods.

Results The numerical examples demonstrated a higher accuracy of the established model than traditional data-driven methods, manifesting a notable 91% decrease in MAPE. This model also displayed reduced dependency on data and alleviated Root Mean Square Error (RMSE) fluctuations by 60% with changes in data size. Moreover, the computational expenses were minimized to a mere 12% of those incurred by the Fluent numerical simulation method. In practical engineering applications, MAE of the suggested framework substantially decreased by 71% and 58% respectively when compared to the Taylor model and the enhanced Taylor model. These findings underscored the effectiveness of the proposed prediction method in resolving the mixed oil concentration distribution in long-distance pipelines.

Conclusion The proposed prediction method of mixed oil concentration in product oil pipelines couples oil mixing progression mechanism constraints with data correction. This method accurately and efficiently predicts mixed oil concentration distribution in long-distance pipelines, guiding the formulation of mixed oil receiving plans for stations and enhancing the intelligent control of oil mixing.