Build of Machine Learning Proxy Model for Prediction of Wax Deposition Rate in Two Phase Flow Water-Oil

Abstract

Wax deposit is one of the major fl ow assurance experienced in the process of oil production and transportation from sub- surface to surface. Large amounts of data are required to perform modeling using existing thermodynamic models such as carbon number data from HGTC. In this paper, a machine learning algorithm using unifi ed model approach from Huang (2008). Two types of input are implemented in order to simulate infl uence of feature selection used in training and testing machine learning which are input A consists of water volume fraction (fw), shear stress (τw), effective viscosity (μe), wax concentration gradient (dC/dT), and temperature gradient (dT/dR) and input B consists of water volume fraction (fw), shear stress (τw), effective viscosity (μe), wax concentration gradient (dC/dT), temperature gradient (dT/dR), shear stripping variable (SV) dan diffusion variable (DV). The random forest with Ntree = 500 known to be the best machine learning method compared to others. Based on accuracy parameter it achieves error parameter R-squared (R2) for training, testing and total data for input A and B are 0.999, 0.992, 0.9975 and 0.999, 0.993, 0.9977, respectively.