Performance of carbon dioxide injection in drained tanks using neural network algorithms
Subject Areas : Petroleum Reservoir Geology
pouya eshaghi
1
,
keivan shayesteh
2
,
mohammad javad khani
3
1 - -1Department of Chemical Engineering
2 - Department of Chemical Engineering
3 -
Keywords: carbon dioxide injection, depleted reservoirs (ROZ), over-harvesting, artificial neural network.,
Abstract :
Injecting carbon dioxide (CO₂) in oil reservoirs is an effective way to increase oil recovery and CO₂ storage. In this study, an artificial neural network (ANN) was used to predict oil recovery and CO₂ storage capacity in depleted reservoirs (ROZ) with respect to geological and well operation uncertainties. Field data from the Smeaheia region, Norway, including 14 key features for optimizing CO₂ injection were identified. Two neural network models, MLP and RBF, were used in this research and their accuracy was evaluated as 91.36% and 94.63%, respectively. In order to optimize the features and reduce the dimensions of the data, the gray wolf algorithm was used, which led to the selection of 10 effective features. These properties included permeability, well pressure, pore volume, compressibility, and porosity-to-height ratio. The optimized models increased the prediction accuracy of CO₂ injection in the MLP model to 97.46% and in the RBF model to 98.97%. These results show that the combination of ANN and optimal feature selection can be a powerful tool for predicting and managing CO₂ injection in oil reservoirs.