Rapid modeling of transient processes in hydrogen-blended natural gas pipelines featuring injection and withdrawal points

Blending hydrogen into natural gas pipelines enables the large-scale delivery and application of hydrogen, contributing to carbon emission reduction and promoting the transition to cleaner energy sources. Accurate and efficient predictions of transient processes in such pipelines are critical for effective management and operation. This study proposes a Fourier Neural Operator (FNO)-based model for the rapid prediction of transient processes in hydrogen-blended pipelines with injection and withdrawal points. The model is trained using numerical simulation data of pressure and flow dynamics, with initial conditions, boundary conditions, and the hydrogen blending ratio as inputs to predict hydraulic parameters along the pipeline over time. By optimizing data preprocessing and neural network hyperparameters, the model achieves high accuracy and efficiency in predicting hydraulic states. The FNO model achieves average Root Mean Squared Errors (RMSEs) of 1.06 × 10 −3 , 1.64 × 10 −3 , and 7.40 × 10 −4 for pressure, flow rate, and density, respectively, with the Coefficient of Determination (R 2 ) exceeding 0.99 for all three variables. Furthermore, the model demonstrates strong generalization capabilities, accurately predicting hydraulic states for pipeline topologies and scenarios beyond the training data. In these cases, the RMSE remains in the order of 10 −3 or smaller. For a 100 km pipeline, the computational efficiency of the FNO model is at least 969 times higher than that of traditional numerical methods , with the maximum efficiency improvement reaching 5168 times as the number of pipeline cases simultaneously predicted increases. These results highlight the potential of the FNO model for real-time applications in hydrogen-blended natural gas pipeline management. • A neural network surrogate model efficiently computes transients in hydrogen-blended pipelines. • A 2D data structure organizes input parameters and simulation results for model training. • The model exhibits robust generalization for different pipeline topologies and temporal resolutions. • The proposed model achieves a significant speedup compared with traditional numerical methods .