Advancing Multiphase Reservoir Simulation: Physics-Guided Expansion from Two-Phase to Three-Phase Datasets

Physics-Guided Data Augmentation (PGDA) refers to a data augmentation approach that integrates physical laws, constraints, or properties of a system to generate new training samples in a machine learning model. Unlike traditional data augmentation techniques that rely purely on statistical transformations, PGDA leverages domain-specific physics based formulations to introduce new data points that adhere to fundamental physical principles governing the system. In this work we propose a physics guided gas estimator to expand a two-phase (oil-water) dataset into a three-phase (oil-water-gas) dataset, ensuring that the new gas phase is realistically incorporated based on fluid dynamics constraints. The augmentations process accounts for reservoir pressure, solubility ratio, and phase equilibrium conditions. This approach successfully extended a dataset to train a machine learning based reservoir simulator.