Abstract
Several multiphase models exist for predicting pressure drops in vertical, inclined and horizontal wells. Most of these models made efforts to unravel the underlying physics of multiphase fluid flow for both mechanistic and analytical models with ranges of applicability. However, these developed models have their specific limitations or ranges of applicability in terms of pressures, rates, basic sediments and water (BSW), choke sizes, fluid ratios, etc thus there is no one model that is universally acclaimed best.
Data from fifteen (15) wells were obtained from a Nigeria field and multivariate linear regression was fitted into the data resulting in a good fit particularly for high production rates/wells. The method of least squares was adopted which involve setting up a set of normal equations, solving them by Gaussian elimination technique and fitting the data into the normal equations. The solution of the primary matrix was obtained through the use of MATLAB software which gave point estimates of the regression parameters leading to the development of multivariate linear regression model. The Bootstrapping method of statistical inference was also used to estimate the confidence interval of these regression parameters, while the Durbin-Watson test statistic was used to test model adequacy. The results of the model compared favourably with existing models within considered range of applicability.
This paper therefore aims to address the challenges of field based multiphase fluid flow model for operational ease using a set of over one thousand production data sets of about forty (40) years of production taken from a renowned field in the Niger Delta. More specifically, our investigations show that taking the infimum (i.e. lower bound) of the parametric bandwidths of the gas rate, average bean size and the model correction factor, b0 the model predictive capability is drastically improved for low to medium rates.
