Fast prediction of coupled oil and environment temperature fields during shutdown of buried waxy crude oil pipelines using neural networks

During the shutdown of buried pipelines carrying hot waxy-rich crude oil, the temperature is likely to drop below the pour point due to heat loss to the surrounding soil environment. This drop can lead to gelation incidents, resulting in significant economic losses. Therefore, in this study, fast prediction models for coupled oil and environment temperature fields during buried pipeline shutdowns are presented, utilizing the Fourier Neural Operator (FNO) network and U-shaped network (UNet). Transient oil and environment temperature fields at the pipeline cross sections are calculated by inputting the shutdown time, the coordinates of the environment temperature field at the pipeline cross section, and boundary conditions. The numerical results are employed to train both the FNO and UNet models. Accurate and fast predictions of oil and environment temperature fields are achieved within 0.5 s for both models, with the FNO model showing slightly better performance in terms of prediction accuracy and efficiency. A root mean square error of 0.015 is maintained for environment temperature predictions, and oil temperature predictions maintain relative errors below 5.0 × 10−4. In four test datasets, the relative prediction errors for oil temperature are kept on the order of 10−3, indicating strong generalization capabilities. Regarding computational efficiency, an acceleration ratio of 1563–2250 is achieved by the UNet model compared to traditional numerical methods, while the FNO model improves this ratio to 2016–2806. These findings offer essential guidelines for the safe shutdown and restart operations of buried wax-rich crude oil pipelines.