Artificial Neural Network Model for Predicting the Viscosity of Crosslinked Polyacrylamide and Polyethylenimine Polymer Gel for Oilfield Water Control

This study addresses a gap in applying machine learning for polymer gel viscosity prediction in improved oil recovery (IOR). A baseline ANN model was developed using 418 experimental datasets from crosslinked PAM/PEI gels under varied conditions, achieving excellent predictive accuracy and demonstrating potential for enhanced oilfield water control.

Repository Description

This repository contains all the underlying data and materials necessary to reproduce the study "Development of an Artificial Neural Network Model for Predicting the Viscosity of Crosslinked Polyacrylamide and Polyethylenimine Polymer Gel for Oilfield Water Control." The work was performed using MATLAB 2022b, and the model was developed with MATLAB’s Neural Network Toolbox.

Included Materials:

• Raw Experimental Data:

All 418 experimental datasets are provided. These datasets include measurements of polymer gel viscosity under varying conditions—shear rate, salinity, ammonium chloride concentration, and silica nanoparticle content.

• Model Input Data and Parameters:

The repository contains the complete variable dataset used as input for the ANN model. This includes detailed descriptions of the input variables, along with the corresponding raw data.

• Model Architecture and Replication Details:

The ANN model is a two-layer feed-forward neural network featuring a single hidden layer with 10 neurons. The model was trained using the Levenberg-Marquardt algorithm. The hidden layer uses the tangent sigmoid (Tansig) activation function, and the output layer employs a linear (Purelin) activation function. Detailed model parameters, including neuron weights and biases, are included to facilitate replication.

• Software and Commands:

MATLAB 2022b was used to develop the model. The “nftool” command from MATLAB’s Neural Network Toolbox was specifically employed to access the Neural Net Fitting app and build the model. All commands, settings, and parameters are documented within the repository.

No data involving human participants were collected, and all experimental data adhered to the relevant institutional and legal standards.

This comprehensive dataset and detailed methodology ensure that other researchers can fully interpret, replicate, and build upon this work in the field of improved oil recovery