ANN-Based Prediction of Laboratory-Scale Performance of CO2-Foam Flooding for Improving Oil Recovery

Abstract

Improving oil recovery by CO2 injection continues to gain momentum in mature oil fields due to its favorable industrial and environmental benefits. One remediation for the poor sweep efficiency of CO2 is co-injection of surfactants to generate CO2-foams in reservoirs. However, it is essential to minimize the expensive and time-consuming experiments required during the laboratory screening of this EOR process for a given reservoir. In this regard, methods to predict RF and Q from reservoir characteristics based on existing laboratory test data are worthwhile. In this paper, we develop the RF and Q prediction models involving optimized multi-layer perceptron (MLP) and radial basis function (RBF) neural networks. These models are applied to a compiled dataset of 214 data records of published CO2-foam injection tests into oil-reservoir cores. The RF and Q prediction derived applying these two models to the compiled dataset are compared. Statistical accuracy measures of the predictions achieved for an independent testing subset (20% of the data records) indicate for RF (MLP: RMSE = 0.0236, R2 = 0.9988; for RBF: RMSE = 0.0197, R2 = 0.9991) and for Q (MLP: RMSE = 0.0283, R2 = 0.9971; for RBF: RMSE = 0.0092, R2 = 0.9991) the excellent prediction performance of the developed networks.