Small-Scale or Large-Scale Machine Learning Proxy Models? a Real Field Case Study

This study explores the potential of machine learning, specifically artificial neural networks (ANN), for proxy modelling in geoscience and reservoir engineering. It highlights the limitations of large-scale ANN models and emphasizes the preference for small, fit-for-purpose proxy models in these fields. The study introduces a step-by-step workflow for constructing efficient small-scale ANN proxy models and demonstrates its practical application in estimating water saturation for a North Sea oil field, achieving high accuracy and generalization through comprehensive feature engineering and data analysis. It is a two-layer ANN with just forty input features estimating the water saturation in all 123,000 active grid cells of the 3D simulation model ten times faster than the numerical flow simulator.