석유회수증진을 위한 pH 민감성 폴리머의 인공신경망 점도 추정 모델 개발

Abstract

Polymer flooding is one of the most common technologies for enhanced oil recovery(EOR) by mobility and conformance control. However, it has technical problems such as high injection pressure with associated pumping cost, creation of unwanted injection well fractures and mechanical degradation of polymers due to high shear near wellbore. Recently the use of partially hydrolyzed polyacrylamide (HPAM) which is a pH-sensitive polymer has been proposed for overcoming the problems of polymer flooding. Since viscosity of pH-sensitive polymer could change up to several times depending on pH and salinity, commercial reservoir simulators cannot estimate its viscosity variations. This limited capability may cause severe errors in polymer flooding designs. Therefore, it is necessary to develop viscosity correlations applicable to pH-sensitive polymer.

HCS(Huh-Choi-Sharma) model which combined modified Brannon-Peppas & Peppas model, modified Mark-Houwink equation, new power-law correlation and Carreau equation was proposed to compute the viscosity of HPAM solution. However, it has some problems such as necessary for empirical parameters and low accuracy in the case of high salinity and in the presence of divalent ions.

In this study, artificial neural network (ANN) which is usually used to model complex and nonlinear relationships between inputs and outputs has been used to estimate viscosity of pH-sensitive polymer. The experimentally measured viscosity data of HPAM and Levenberg-Marquardt algorithm is used to train ANN viscosity model which is composed of the four layers, one input layer with six neurons containing pH, polymer concentration, degree of hydrolysis, molecular weight, salinity, and shear rate, one output layer with one neuron which is viscosity, two hidden layers with five neurons each.

As a result of this study, ANN viscosity model which has only 6 input parameters without empirical parameter can estimate viscosity more precisely than HCS model which needs 21 parameters especially in high salinity condition. Unique features of ANN which are adaptive learning capabilities and expansibility could make ANN viscosity model to estimate viscosity more accurately and to be applicable to various conditions if sufficient data are available for training. ANN viscosity model suggested in this study could be implemented in an EOR process simulator for optimal mobility control applications.