Fluid Production Prediction using Machine Learning in the Lower T of the Sacha Field

Abstract

The purpose of this study is to predict the fluid production of the wells drilled during the year 2021 in the lower T sandstone in the Sacha field. The petrophysical and fluid information was considered for the construction of the model from a standard well using commercial software, which provides the analysis of well behavior. In parallel, using the Python programming language through Machine Learning, two algorithms were developed: one based on the pump intake pressure (PIP) data of the electric submersible pump (ESP), and another with PIP data and salinity of the reservoir formation water. The prediction of fluid production with respect to the real production, found an error of 2% with the commercial software while in the two simulations through Python an error of 10% and 0.5% was obtained respectively. In the case of gas production prediction, the real value is 0.07 MMSCFD, while the one obtained by simulation with commercial software is 0.41 MMSCFD. For the case of the first and second simulation with Python, a better approximation of 0.11 MMSCFD and 0.10 MMSCFD was obtained respectively. The increase of variables in Python allows the reduction of the percentage of error and increases the adjustment of the of fluid and gas production prediction, in this case the PIP of the BES and salinity of formation water.