The evaluating of the pressure drop due to multiphase flow in vertical pipes and inclined pipes is crucial for oil and gas industry. Numerous correlations and models have been developed to calculate the pressure drop in vertical wells but the effectiveness of these models is still under debate. Most of the correlations and models were developed to calculate the pressure drop due to multiphase flow based on accurately and reliably measured flow parameters. However, they can work best within the proposed data ranges. Their accuracy degrades if they are used for data out of the measured ranges.
Artificial Intelligence (AI) has proven to be an alternative solution to many of the problems where physics and classic statistics fail to provide satisfactory solutions due to limiting assumptions and complicated reality. Different AI methods, viz., Fuzzy Logic (ANFIS), Neural Networks (ANN), Support Vector Machine (SVM), and Decision tree (DT) were used in predicting the multiphase flow pressure drop in surface pipeline and production tubing for oil fields. 239 published real field datasets were used in constructing the flow line model and about 795 datasets were used to construct the pressure drop from the well head pressure to the bottom hole flowing pressure (tubing model).
The models were tested by dividing the data into three categories:, (60%) were used for training and (15%) for validation, and (25%) for testing. The results of testing showed that ANN, ANFIS, and SVM give better predictions than the common correlations in case of pressure drop from the wellhead to the GOSP (flow line) or from the wellhead to the bottom hole location (tubing). The average absolute percentage error (AAPE) was 1.15% for ANN, 0.39% for ANFIS, and 9% for SVM, and 77% for DT for the flow line model and the error was 3.8% for ANN, 4.5% for ANFIS, 4.3% for SVM, and 3% for DT for the tubing model.