Oil-Water Relative Permeability Prediction Using Machine Learning

Abstract Relative permeability is one of the most significant reservoir characteristics in the petroleum industry. It captures the fluids behavior inside the porous space within the reservoir. It considers the effective permeability of the fluids in the reservoir which ultimately lead to the understanding of the fluid behavior inside the pores. Also, using relative permeability curves, we can estimate the reservoir's oil or gas recovery. Furthermore, enhanced oil recovery techniques utilize relative permeability curves to evaluate their performance. The well-known practice to develop any relative permeability curve is by conducting core flooding experiments which are relatively time consuming especially if it is needed to be done on several wells with different core samples. Also, it would be costly data set to acquire since it requires special lab sets and conditions. Time and cost are the main factors making relative permeability a very hard to obtain information for any reservoir. Several models and empirical relations have been built to calculate and present relative permeability without going through the lab experiments, each model has its uncertainty. This paper captures the approach to predict relative permeability curves (oil and water) from a set of data collected from one reservoir using machine learning. Data used is generated from special core analysis lab experiments (core flooding) of unsteady state oil and water relative permeability. Core flooding experiments represents several water saturations at which the core been flooded to, at every water saturation a water and an oil relative permeability value is obtained. To represent the reservoir efficiently and addressing several aspects of its 56 relative permeability curves (from 56 composites) have been collected from different wells in the same reservoir. Adding up to a total of more than 7,000 data sets (different water saturations). Two models have been built, one for predicting the relative permeability of oil at several water saturations and the second model is for the relative permeability of water. Main input data are water saturations, connate water saturation, residual oil saturation, porosity, oil viscosity, water viscosity, (several basic core properties) and wettability. The outcome for each model is one, either oil or water relative permeability. The main added value of this work is creating a workflow and models to predict water and oil relative permeability using main reservoir data with high accuracy and without conducting any special core analysis.