Oil-Gas Multiphase Flow Surrogate Model Embedded With Mechanism Formulas

Abstract Accurate predictions of pressure and temperature drops in oil-gas multiphase flow pipelines are vital for the design and operational efficiency of oil and gas field pipelines. Mechanistic models, though precise, require lengthy computations, rendering them impractical for the extensive calculations needed during optimization. Moreover, their accuracy is often hampered by limited experimental data ranges. Machine learning surrogate models emerge as a solution, offering swift calculations with high accuracy, yet they typically overlook flow-specific PVT properties and gas velocity. Additionally, the “black box” nature of machine learning is often criticized. Addressing these gaps, our research presents a deep neural network surrogate model that offers faster computation speeds while maintaining nearly the same level of accuracy as mechanistic models. This model synergizes with mechanism formulas, including momentum conservation, and encompassing flow condition PVT properties. It adeptly predicts pressure and temperature changes in pipelines of any orientation, with input parameters including liquid flow rate, gas-oil ratio, inlet pressure/temperature, pipe length, diameter, inclination, and gas-liquid densities (at standard conditions). The outputs are outlet pressure/temperature, gas phase density (at a specific temperature and pressure), gas phase superficial velocity, liquid holdup, and two-phase friction factor. Training with the PIPESIM simulator data, it demonstrates a remarkable prediction accuracy of over 98%, significantly surpassing traditional machine learning and neural network approaches. Additionally, it boasts a computational speed three orders of magnitude faster than conventional mechanistic models, proving its efficacy for extensive, rapid calculations necessary for large-scale pipeline optimization.