Prediction of Minimum Miscibility Pressure (MMP) of CO2-Crude Oil System Based on GWO-RBF Neural Network

Accurately evaluating the minimum miscibility pressure (MMP) of CO2-crude oil systems is a critical problem in the design of CO2 miscible flooding of reservoirs. Based on optimization algorithm and artificial intelligence model, a novel model integrating grey wolf optimizer (GWO) algorithm and radial basis function (RBF) neural network is built through measuredMMPdata fitting, which is named GWO-RBF. On the basis of Spearman rank correlation analysis, the proposed model selects reservoir temperature, oil C5+ molecular weight and mole fraction of intermediate oil components (C-2-C-4, CO2, H2S) as the independent variables. The GWO-RBF model parameters (base function width s, hidden layer node number M) are optimized using 40 sets of published CO2-crude oil system MMP data. Prediction is performed on extra 10 sets of non-training MMP data. The results show that the average absolute relative deviation (AARD) of the GWORBF model is 2.73% on training sets, while the AARD of model prediction is only 3.11%. On training and prediction sets, the maximum absolute relative deviations (ARDs) are 8.30% and 9.99%, respectively, while the minimum ARDs are 0.19% and 0.00%.