Developing a Generalized Scaling-Law for Oil Recovery Using Machine Learning Techniques

The universal dimensionless time is an important parameter to predict a real oil field behavior from a scaled laboratory experiment. The dimensionless time uses a group of parameters that are inherited to the properties of both fluids and rocks of the reservoirs. In addition to the injection velocity which dynamics of the process depends on. In this work, we proposed a new power-law scaling velocity as a function of characteristic injection velocities, which can be linked into the dimensionless time. Based on machine learning techniques, namely, k-nearest neighbor algorithm (k-NN), artificial neural networks (ANNs), support vector machine (SVM), and random forest (RF) for experimental oil recovery data and real time is employed to estimate the dimensionless time of oil recovery. It was found the SVM method is the best machine learning technique to predict dimensionless time-scale based on the primary physical data of the oil and rock.