Predictive Modeling of two-phase choke behavior in one of the Iranian oil fields using machine learning

Production engineers play a crucial role in maintaining reservoir productivity and ensuring it stays at an optimal level throughout the production period by preventing excessive production and managing flow through wellhead chokes. Wellhead chokes are devices installed in flowing pipes to resist pressure, regulate production rates, prevent water and gas coning, and control pressure to keep wellhead equipment functioning properly. These chokes are categorized into two main types: fixed (positive) chokes and adjustable (variable) chokes. The flow rate passing through wellhead chokes depends on factors such as wellhead pressure, choke diameter, and liquid production rate. In this study, a dataset of ۵۶۵ pressure measurements along with liquid flow rate, choke diameter, and gas-liquid ratio was analyzed. The study compared two empirical formulas by Gilbert and Ross to machine learning models combined with optimization algorithms. The Nelder-Mead optimization method yielded the highest accuracy, outperforming other algorithms as well as the empirical formulas.

Keywords:

genetic algorithm , machine learning , choke modeling , MLP neural network , Nelder-Mead