New approach to evaluate the performance of highly deviated water injection wells using artificial neural network

Abstract

Water injection wells are used to improve the hydrocarbon recovery by maintaining the reservoir pressure and displacing the remaining oil into the production wells. Water injection wells can be vertical, deviated or horizontal wells. Highly deviated wells are used to improve the well injectivity and reduce the injection pressure. Several models were developed to determine the well injectivity for vertical or horizontal injection wells. However, few models are reported to estimate the performance of slanted or highly deviated wells. This study introduces a new approach to evaluate the performance of highly deviated wells used for the water injection applications.In this work, the well injectivity was determined utilizing the reservoir properties and the well geometry. Artificial neural network (ANN) was used to evaluate the injectivity index based on the reservoir thickness, formation permeability and well inclination. More than 550 cases were used to train and test the intelligent model. Two approaches were utilized to estimate the well injectivity; direct and indirect approaches. The reservoir thickness, formation permeability and well inclination were used as input parameters in the direct approach. While, in the indirect approach, all input parameters were combined into one input and then the well injectivity was determined using a single input parameter. Moreover, an empirical correlation was extracted utilizing the optimized ANN model. The extracted equation was verified using validation data, an average absolute difference (AAD) of 1.22 (STB/D/psi) was achieved indicating the effective performance of the developed ANN-based equation.The novelty of this work is that an effective approach is presented to evaluate the injectivity of highly deviated wells. The presented approach proved its reliability in predicting the well injectivity for heterogenous reservoirs. For the first time, an empirical correlation is extracted from the ANN model to estimate the injectivity index. The proposed ANN-based equation outperforms the popular models for determining the performance of highly deviated injection wells. The developed equation allows an accurate, easy and direct determination for the well injectivity for the highly deviated wells.