Estimation of flowing bottom-hole pressure in multiphase flow is done through empirical correlations like Beggs and Brill or Gray correlation. These correlations were developed in time when computing power was limited and some were developed based on water-air mixtures, making them less accurate in real oil and gas production scenarios. Regardless, due to lack of other alternatives, general empirical correlations, like the ones mentioned previously, are widely used in the industry for estimating flowing bottom-hole pressure in multiphase flow.
Machine learning and Artificial Intelligence (AI) are emerging techniques in analyzing a large set of data to identify relations and patterns in multivariate problems. The method used in this paper – Quasi-Monte Carlo (Latin Hypercube sampling) – randomly selects input data to create empirical correlations between flowing parameters and bottom hole pressure. In this study, production data (oil, gas, water flow rates, and flowing wellhead pressure) from an oil well is used as input, and an algorithm based on Latin Hypercube sampling selects random variables from each of the parameters to generate several correlations. The correlation with the minimum error is chosen.
From the developed equation, the calculated flowing bottom hole pressure was within an average error of 5%. The advantage of this method is that it does not require any tubular information, such as tubing internal diameter, surface roughness, or tubing restrictions. In addition, the average reservoir pressure can be estimated from the constant of the model. Furthermore, the application of this method can be used to indirectly infer several fluids or rock properties.
This is a novel method to statistically estimate flowing bottom-hole pressure using only production data. This method can also be expanded to gain knowledge about other reservoir properties.
