Application of machine learning techniques to predict viscosity of polymer solutions for enhanced oil recovery

Polymer flooding has become one of the most developed and implemented enhanced oil recovery (EOR) techniques. The principal controlling factor in polymer flooding is the viscosity of the polymer solution which helps to lower the mobility ratio and improve sweep efficiency. However, designing a polymer solution’s viscosity is both a time and resource-intensive task as several parameters need to be designed to maintain the desired polymer viscosity such as brine salinity, polymer concentration, temperature, etc. This study aims to find a quick and accurate method to determine the viscosity of three modified hydrolyzed polyacrylamide (HPAM) based polymers namely DPTLB-2070, SAV-10, and SAV-333 polymers as a function of the critical parameters of shear rate, polymer concentration, and temperature. Four different data analysis techniques have been applied including multiple linear regression (MLR), support vector machine (SVM), regression decision tree (RDT), and artificial neural network (ANN). The results show that MLR is not suitable for predicting polymer viscosity because of the nonlinearity of the problem. Among the machine learning methods, the ANN model having two hidden layers and five neurons in each layer has provided acceptable results with a correlation coefficient of 0.99 for training, validation, and testing datasets in the case of all three polymers.