New polymer rheology models based on machine learning




Alqahtani, Abdulwahab Saeed

Journal Title

Journal ISSN

Volume Title



A successful polymer-type EOR project relies upon many factors, including an adequate characterization, description, and prediction of the polymer’s rheology. A high polymer viscosity can improve the mobility and sweep efficiency, but can also lead to poor injectivity. Polymers are generally non-Newtonian and the rheology is a function of in-situ shear rate, polymer concentration, salinity, temperature, molecular weight, and molecular structure. A priori estimation of polymer rheology using models is important for design of polymer floods and prediction using numerical reservoir simulators. Existing models require many fitting parameters, are purely empirical, and can rarely be used for a priori estimation. The objective of this work was to develop new models to predict the viscosity of HPAM polymers used in enhanced oil recovery (EOR) and implement them into a chemical flooding numerical reservoir simulator. The study uses a combination of fundamental, physical models and machine learning methods to develop new predictive models. The data used in the study includes the measured polymer rheology at various polymer concentrations, molecular weights and types, temperatures, and brine salinity and hardness. Data are first fit to the 4-parameter Carreau’s model and then advanced machine learning techniques are used to develop the models of the Carreau parameters with the aforementioned solution properties. The models are then used to predict the rheology of new samples which are validated against data measured on an ARES G2 rheometer. All data fit the 4-parameter Carreau model well. The new models for the zero-shear viscosity, shear thinning index, and time constant are a function of temperature, polymer concentration, salinity, hardness, and molecular weight using less than ten parameters


LCSH Subject Headings