An Integrated Genetic-Algorithm/Artificial-Neural-Network Approach for Steady-State Modeling of Two-Phase Pressure Drop in Pipes

Abstract

Modeling of multiphase flow represents the cornerstone of oil/gas-production systems. Accurate pressure-drop estimation is crucial in the design and operations of subsea architectures. However, the complexity of the underlying physics governing the transport of mass, momentum, and energy considerably limits the accuracy of the current state-of-the-art models. In this paper, we resort to artificial intelligence to develop a unifying artificial-neural-network (ANN) model encompassing all flow conditions. A genetic algorithm (GA) is used to find the optimal input combination from a broad pool of candidates leading to the best prediction accuracy. To train and validate the model, we used the Stanford multiphase-flow database (SMFD). Comprising 5,659 measurements (1,800 of which are actual field data), the SMFD is the largest of its kind encompassing several published data sets. Eighty percent of the data were used to train the model (4,527 measurements) and the remaining 20% (1,132 measurements) were used for validation. The proposed model was compared with two published models, the Beggs and Brill (1973) model, which is widely used in the oil and gas industry, and the Petalas and Aziz (2000) model (a preeminent mechanistic model). The proposed model was proved to significantly increase the prediction accuracy across all pipe-inclination ranges (up to 88%) and also all observed flow patterns (up to 71%). This is a major contribution with potential benefits to the oil and gas industry, where, because of the limited accuracy of the current models, much conservatism is used in the design of subsea architectures, leading to shortfalls of millions in profits.