A novel empirical correlation for waterflooding performance prediction in stratified reservoirs using artificial intelligence

Water has been used as an injected fluid for decades to improve oil recovery, commonly known as waterflooding. Simulating this process is very expensive, especially for the post-water breakthrough analysis in stratified oil reservoirs. The existing correlations do not predict waterflooding performance in heterogeneous reservoirs accurately. Most of the methods do not account for pattern flooding and consider piston-like displacement with non-communicative layers. In this study, a model has been developed using artificial neural networks (ANNs) for predicting the recovery performance of a layered reservoir undergoing a five-spot-pattern waterflood. In addition to the ANN model, a mathematical equation is presented based on ANN to predict the oil recovery in pattern waterflooding with and without crossflow between the layers for different rock wettabilities. A novel parameter—wettability indicator (WI)—has also been introduced that can be used to quantify the rock’s wettability based only on the relative permeability curves. The results showed that the introduction of the new term (WI) significantly decreased the simulation runs in comparison with existing relative permeability models. ANN approach was compared with non-linear regression (NLR) and adaptive neuro-fuzzy inference system (ANFIS). The ANN model outperformed NRL and ANFIS in terms of least mean absolute percentage error (MAPE) and highest coefficient of determination (R²). The new correlation was tested with an unseen data set, two different real field cases, an analytical model, and a semi-analytical model. The training and testing data show good match and accuracy with R² of 0.9973 and 0.997, respectively. MAPE of the predicted recovery efficiency using a blind data set was around 7%. The developed correlation can be a useful tool for a quick estimate of the waterflood oil recovery before a large simulation model is built and ran.