Rapid Prediction Models for Oil Temperature and Surrounding Environment Temperature Fields in Buried Hot Crude Oil Pipelines

Abstract

Wax-rich or high-viscosity crude oil requires heating for long-distance transportation in buried pipelines. Monitoring the temperature distributions and variations of both oil and surrounding environment is crucial to ensuring safety and economic efficiency. This study proposes two models for rapidly predicting the steady-state oil temperature and surrounding environment temperature fields based on the Fourier neural operator (FNO) network and U-shaped network (UNet), respectively. These models leverage numerical results as training data, incorporating boundary conditions and environment grid coordinates at the pipeline cross section as inputs to predict the temperature distributions of both oil and the surrounding environment along the buried pipeline. With optimized hyperparameters, the models achieve accurate and efficient predictions. The FNO and UNet models had average RMS errors (RMSEs) in environment temperature field prediction of 2.68×10−3 and 5.49×10−3 at the pipeline cross section, respectively. For oil temperature predictions, the FNO model had an average relative error of 1.49×10−4, compared with 2.55×10−4 for the UNet model, with average absolute error values of 5.32×10−3 and 7.24×10−3, respectively. Moreover, both models exhibited strong generalization, with an average RMSE in the environment temperature field prediction of less than 3.5×10−2 and an average relative error in oil temperature predictions of less than 2.1×10−3 across different data sets. Comparatively, the FNO and UNet model had slightly higher prediction accuracy than the UNet model. In terms of computational efficiency for a 100-km pipeline, these models offer improvements of at least 116.25× over the numerical simulation method, with a maximum improvement of 1,775.03× as the number of simultaneously predicted pipeline cross sections increases.