Application of a Hybrid Numerical Reservoir Simulation and Artificial Neural Network for Evaluating Reservoir Performance Under Waterflooding

Abstract

In the current study, artificial neural networks (ANNs) and numerical reservoir simulation (NRS) are applied to produce a NRS-ANNs hybrid model, for the analysis of reservoir performance prediction under waterflooding of a ZH86 block in the Zhaozhouqiao oilfield, China. Five input data sets were extracted from the history-matched model and utilised to build the proposed model. A 5-10-10-6-6-1 NRS-ANNs hybrid model architecture was selected to analyse the performance of reservoir prediction based on a minimal root mean square error (RMSE) of the testing data sets attained. For validation data sets, the prediction performance of the selected NRS-ANNs hybrid model had the minimal RMSE of 0.0274 and the coefficient of determination (R2) and coefficient of correlation (CoC) values of about 0.9999 for observed oil production data. Furthermore, a study investigated the correlation between the five input variables, block liquid production rate (BLPR), block water production rate (BWPR), block water cut (BWCT), block water injection rate (BWIR) and block reservoir pressure (BRP) and the output variable, simulated oil production rate (SOPR). The study found a positive correlation between the SOPR and the three input variables BLPR, BWIR, and BWCT. However, there was a negative correlation between SOPR and the two input variables, BRP and BWPR. Meanwhile, a study found that segment B experienced a 3.8% increase in the BLPR input variable, while segments A and C showed declines of 1.3% and 1.6%, respectively. These variations in liquid production rates corresponded to alterations in the SOPR of 4.3%, 1.9%, and 9.7% for segments A, B, and C respectively. Moreover, the NRS-ANN hybrid model prediction performance was compared with NRS model results. The accuracy of the NRS-ANN hybrid model in predicting SOPR was found to be more than 1125 times greater than that of the NRS model. Based on to the results, it can be concluded that the proposed NRS-ANN hybrid model is an accurate and useful tool for analysing the performance of the reservoir under the waterflooding oil recovery technique.

Keywords: Artificial Neural Networks, Numerical Reservoir Simulation, Waterflooding, Feed-forward Backpropagation, Performance Prediction.