predictive data analytics application for production optimization in oil reservoirs

The oil recovery factor (rf) is the most important parameter for all exploration and production (e&p) companies, particularly during the early reservoir life, because numerous investment decisions are made based on the amount of hydrocarbon that can be extracted from the target asset by using the available techniques and operational practices. the estimation of this parameter could be achieved through several techniques, but the accuracy of these methods depends on specific data availability, which is strongly dependent on the reservoir age. most of the presented intelligent models have only been developed for the prediction of the oil recovery factor in sandstone reservoirs. in this study, the top-down modeling approach was used for oil recovery factor prediction as a function of rock type, production mechanism, porosity, permeability, connate water saturation, reservoir temperature, api gravity, original reservoir pressure, and stooip for both sandstone and carbonate reservoirs under either water-drive or solution gas-drive production mechanisms, by implementing different types of neural network algorithms (multilayer perceptron (mlp) and cascade neural network (cnn)). in addition, the mlp and cnn models were trained using different training algorithms, namely, levenberg-marquardt algorithm (lma) and bayesian regularization (br). the results exhibited the outperformance of the mlp model trained with the bayesian regularization algorithm for forecasting the rf with an rmse of 6.3283, coefficient of determination of 0.8681, and correlation coefficient of 0.9317 for the total dataset. the developed intelligent models based on tdm can be applied for asset evaluation, selection of suitable production optimization strategies, and helping e&p companies for further field development planning.