Diagnosis of Water Production using Artificial Neural Network Models

Abstract

A data-driven approach to solving the problem of excess water production was the focus of this study. In this research, several reservoir models were simulated for 26 years, and data obtained after simulation was used to develop two artificial neural network models to predict the water cut for various producers. The first 22 years of data obtained from simulation was used to train the first neural network model, and the model was then used to forecast the water cut for the last four years of production. The developed neural network model was able to accurately forecast the water cut for the last four years of production of a well, with a regression coefficient of 0.9985 and mean square error of 1.5985*10- 5 . The second neural network model developed served a slightly different purpose. The model was developed to investigate its sensitivity to changes in reservoir conditions. Data used for training the second neural network model was obtained from the simulation results of a set of reservoir models, and the model was used for predicting well water cut of producers in a slightly modified version of those reservoir models. To investigate the sensitivity of the second model, the modified parameters included; permeability contrast, horizontal permeability, perforation interval of producers, the separation distance between an injector and a producer, and the initial aquifer pressure. The prediction by the second model was as good as the forecast by the first model and even better in some cases, indicating that the data- driven approach proposed in this work can serve as a technique for tackling the problem of excess water production in the oil and gas industry.