Promises of Artificial Intelligence Techniques in Reducing Errors in Complex Flow and Pressure Losses Calculations in Multiphase Fluid Flow in Oil Wells

Abstract

Abstract Empirical correlations normally used to calculate pressure losses in vertical and horizontal pipes involve complex calculations that normally rely on estimation of other complex parameters such as liquid hold up and flow regimes before arriving at values of pressure losses. Errors in estimation of hold up and flow regimes using empirical correlations propagate to error in pressure loss calculation. Hence, available empirical correlations perform pressure loss calculations with certain degree of errors which can translate to poor design of production systems. Our objective is therefore to use new techniques-artificial intelligence techniques- to calculate pressure losses during natural flow of multiphase fluid through tubing in vertical wells. The method uses some real multiphase fluid flow data (such as pressure losses, flow and fluid properties) with which an artificial intelligence is developed and trained to understand the relationship between pressure losses with fluid and flow properties. The artificial intelligence is subsequently used to perform similar calculations under new fluid properties and flowing conditions. In the literature, many papers have shown that artificial neural network (ANN) technique can estimate multiphase pressure loses more accurately than some selected empirical models. However, to the best of our knowledge, none or few papers have shown the applicability of support vector machine which is another powerful AI technique. In this paper, we applied two AI techniques-artificial neural network (ANN) and support vector machine (SVM) to predict bottom-hole pressure in a multiphase flow well using field data and we then compared results with some commonly used empirical correlations commonly used in the industry using graphical and statistical error analysis. The empirical correlations used are: Duns and Ros; Hagedorn and Brown; Fancher and Brown; Mukherjee and Brill; Beggs and Brill; Orkizewski; and Petroleum experts II. Our results showed that the two AI methods predicted bottom-hole pressure with accuracies higher than the empirical models. The correlations of coefficient and results error analysis for all the models are presented in tabular and graphical forms for ease of comparison. Finally, it is worth mentioning that artificial intelligence is an emerging technology currently used in the petroleum industry to solve complex engineering problems and its application in multiphase pressure calculations is promising as shown in this paper.