Machine learning-based model for phase equilibrium calculation in multi-component reservoir simulation

摘要

Techniques related to learning a surrogate model based on machine learning for phase equilibrium calculations are disclosed. In one practice, each equation of state (EOS) of one or more regions of the reservoir is made for a sample of downhaul fluid obtained from one or more regions of the reservoir. It is identified based on the results of multiple pressure, volume, or temperature (PVT) tests. The composition of the downhaul fluid sample is identified and spatially mapped based on interpolation between one or more regions of the reservoir. Using the identified EOS, testing of one or more PVTs against the spatially mapped composition of the downhaul fluid is simulated and a reservoir composition database is created. One or more machine learning algorithms are trained using the composition database, phase stability is predicted using the trained machine learning algorithms, and flash calculations for multi-component reservoir simulation. To execute.