Performance Prediction of Waterflooding in Western Canadian Heavy Oil Reservoirs Using Artificial Neural Network

Abstract

This study presents a successful application of multivariate partial least-squares (PLS), response surface methodology (RSM), and artificial neural network (ANN) to develop a new diagnostic tool for performance prediction of waterflooding in heavy oil reservoirs. The data used in this study consist of 120 operational and reservoir parameters for 177 waterfloods in heavy and medium oil reservoirs in western Canada (i.e., Alberta and Saskatchewan). This study also used 15 numerically devised indices for performance evaluation of water injection, based on collected injection and production histories of studied waterfloods. To reduce the number and complexity of input parameters of ANN models, a comprehensive PLS analysis was conducted. In addition to parameter selection, the PLS model provided a more in-depth understanding of differences between heavy and medium oil waterfloods. Next, RSM was used to improve the quality of the database for a selected combination of 38 reservoir and operational parameters by predicting some of the missing data. Finally, ANN models were developed using the feed-forward backpropagation algorithm with momentum for error minimization. The developed models show the superior ability of the ANN for creation of an efficient reservoir engineering tool for fast performance prediction of waterfloods using easily obtainable operational and reservoir parameters. The developed models in this study can be incorporated into reservoir engineering, risk assessment, and production optimization software programs to improve the quality of predictions based on more than 50 years of waterflooding experience in western Canada.

A priori error estimation of upscaled coarse grids for water‐flooding process

“…Therefore, it can be stated that single-phase pressure and velocity fields can be close approximations of two-phase pressure and total velocity fields under similar boundary conditions. In fact, they are exactly the same for the unit mobility ratio displacement processes (l tot ¼ 1) in which Equations (11,14) as well as Equations (12, 16) become similar. Now let us focus on the saturation equation, Equation (15).…”

“…Comparing single-phase and two-phase flow pressure equations, Equations (11,14), and single-phase and total velocity equations, Equations (12,16), one may note their significant analogy. Therefore, it can be stated that single-phase pressure and velocity fields can be close approximations of two-phase pressure and total velocity fields under similar boundary conditions.…”

“…To this end, fluids were considered incompressible and the effects of gravity and capillary forces were assumed negligible. As a result, the governing equations of the water-flooding process can be stated by Equations (14,15). These equations were solved using the finite difference method employing the IMPES [30] (Implicit Pressure Explicit Saturation) scheme in which pressure is solved implicitly while the saturation field in each time step is calculated explicitly.…”

Prediction of Polymer Flooding Performance with an Artificial Neural Network: A Two-Polymer-Slug Case

“…To assess the fitness and prediction To reduce data redundancy, avoid the effects of magnitude differences, and improve the integrity of the training performance of the ANN model, the simulated data were normalized. Ahmadloo et al [26] define data normalization as the process of standardizing the possible range of the input data. In our study, the possible range was set by constant values falling between 0.1 and 1.1 obtained after calculation using:…”

Artificial Neural Network Model for Alkali-Surfactant-Polymer Flooding in Viscous Oil Reservoirs: Generation and Application

“…The weights and biases will be updated throughout the training process; principally, a more accurate network can be obtained with a larger volume of data for training [39,40]. On the other hand, as the use of an inadequate number of data for training can inevitably cause overfitting, which provides very high accuracy for estimating the training data but fails to perform in different data sets.…”