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

2017 Energies  
Many previous contributions to methods of forecasting the performance of polymer flooding using artificial neural networks (ANNs) have been made by numerous researchers previously. In most of those forecasting cases, only a single polymer slug was employed to meet the objective of the study. The intent of this manuscript is to propose an efficient recovery factor prediction tool at different injection stages of two polymer slugs during polymer flooding using an ANN. In this regard, a
more » ... egard, a back-propagation algorithm was coupled with six input parameters to predict three output parameters via a hidden layer composed of 10 neurons. Evaluation of the ANN model performance was made with multiple linear regression. With an acceptable correlation coefficient, the proposed ANN tool was able to predict the recovery factor with errors of <1%. In addition, to understand the influence of each parameter on the output parameters, a sensitivity analysis was applied to the input parameters. The results showed less impact from the second polymer concentration, owing to changes in permeability after the injection of the first polymer slug. the aqueous phase. However, the recovery of incremental oil compared to waterflooding represents an economic incentive for applying polymer flooding. Recovering oil using this method is a challenging task if the water-to-polymer mobility ratio is unfavorable. To avoid such a situation, assessment of the performance of polymer flooding must be conducted to predict the recovery factor or cumulative production resulting from the operation. Conventional methods have been used to evaluate the performance of polymer flooding by predicting the recovery factor and/or the cumulative production, the bearing on enhanced oil recovery's (EOR) efficiency on the effects of capillary pressure, polymer on viscosity, interfacial tensions, and the wave structures associated in two space dimensions resulting in fingering; these include the use of reservoir numerical simulations [9-11] fractional flow theory [12, 13] , and mathematical methods [14, 15] . These methods have numerous drawbacks, such as the requirement of a substantial amount of data related to the geology and geometry of the reservoir, or the fluids and rock properties. The processing of this data is a time-consuming operation, resulting in inaccurate results owing to multiple errors. Developing a simple, fast, and accurate prediction tool to evaluate polymer flooding performance is strongly needed. There are numerous studies available in the literature where artificial neural networks workflow has been developed to tackle problems of different nature in petroleum engineering. Based on the ability of ANNs in solving identification problems, Masoudi et al. [16] were able to determine the net pay zones of two reservoirs, a carbonate reservoir of Mishrif and a sandy reservoir of Burgan in Iran, obtaining a classification correctness rate >85%. Despite the large abundance of wireline logs in most drilled oilfield wells, the core data essential to the determination of water saturation are only available in few wells. Because of that lack, Al-Bulushi et al. [17] designed an ANN which with a R 2 of 0.91 and a root mean square error (RMSE) of 2.5 was able to predict water saturation by later applying sensitivity analyses to confirm the robustness of the model used. ANNs have also been applied in prediction of oil recovery factors and cumulative oil production. By using higher-order neural networks (HONNs), Chithra Chakra et al. [18] were able to predict petroleum oil production without requiring sufficient training data. Mohammadi et al. [19] however put the emphasis of their studies on the prediction of oil recovery factors in CO 2 injection. The result was mainly appreciable regarding the RMSE, which was evaluated to be 0.396%. For EOR processes and especially chemical enhanced oil recovery (CEOR), a number of studies have been dedicated to the utilization of ANN. By applying ANN to the viscosity estimator of Flopaam™ 3330S, Flopaam™ 3630S and Kang et al. [20] came to the conclusion that ANN model has higher polymer viscosity prediction accuracy compared to the conventional prediction model known as the Carreau model. Al-Dousari et al. predicted the recovery factor of surfactant polymer flooding. In the first case, the prediction was made at three different pore volumes (0.75, 1.5, 2.25) by applying a blind-test on 125 data sets, which gave an average absolute error of 3% [21] . In this study, we predicted the recovery factor (RF) during polymer flooding using a neural network at three different periods: after waterflooding (RF 1 ), after the injection of the first polymer slug (RF 2 ), and after the injection of the second polymer slug (RF 3 ). The results of this study can serve as an example to show the capacity of an ANN in predicting the performance of the injection of two or more polymer slugs during polymer flooding. Furthermore, a sensitivity analysis can be applied to the input parameters for finding the best-performing parameters to maximize the recovery factor. Methodology Data Gathering and Polymer Flooding Simulation Reservoir data and polymer concentrations were gathered from previous attempted studies in which polymer flooding had been simulated [10, 22] . Reservoir simulations coupled with polymer concentrations were conducted to obtain a bank of polymer flooding data. The study consisted of a sandstone reservoir with a five-spot well pattern composed of four producer wells and one injector well discretized into 30 × 30 × 5 grid blocks. The producer wells were located
doi:10.3390/en10070844 fatcat:ee2se7j2ozd2bexcz47zgsa5bi