Medição de vazão virtual: avaliação de desempenho de simulação fenomenológica e de rede neural em estudo de caso na província do pré-sal

Abstract

Typically, a produced fluid from an oil well is a multiphase mixture of oil, gas, water and solids. For the correct measurement of volumes, it is required to split into single phases in a test separator. An alternative would be applying a Multiphase Flow Meter (MPFM), but it is not always technically possible and economically viable. Virtual Flow Metering (VFM) is another great alternative, especially in applications with many wells, where numerical models calculated the flow rates of the oil, gas and water phases, from simple field data, such as pressure and temperature. The first principles VFM using multiphase flow simulator and multilayer perceptron (MLP) artificial neural network (ANN) using the Scikit-Learn library for Machine Learning in Python, showed promising results when applied to a dataset from a Pre-salt oil well. The results were compared with test separator measuring, in 325 days of operation, with maximum errors of approximately 10%, but better results when applied model adjustment. This is consistent with the references, and highlighting it is a topic in vigorous technical progress.