New approach to Estimate Oil Recovery Factor for Water Drive Sandstones Reservoirs through Applications of Machine Learning

Abstract

Mature heavy oilfields in the Northern Peruvian Jungle have produced oil for over 40 years under waterdrive mechanism, with a wide range of ultimate recovery factor in between 10% to 60%; a reasonable estimation of recovery factor at an early stage of development and/or production is quite critical for sizing and scheduling development plans, as well as CAPEX investments. This research article introduces a new approach that integrates empirical correlations, analytical methods and machine learning algorithms to estimate oil recovery factor in water-drive sandstone reservoirs at early development and production. Preliminary studies showed that the most representative reservoir and fluid parameters, such as reservoir size, porosity, permeability, net thickness, residual oil saturation, API gravity, oil viscosity and initial pressure, which are typically measured during the exploration, appraisal and early development stages, can be correlated to expected recovery factors obtained from mature fields with long production history. Literature empirical correlations will be initially tested with available information of oilfields of Marañón Basin to estimate correlation coefficient. Different regression ML learning algorithms will be compared using existing data to select the best one providing the most precise predictions. A new empirical correlation that significantly outperforms traditional industry equations will be adjusted with ML algorithms weights and biases. The results of this comprehensive study will contribute to a better understanding of the water drive mechanism in the oilfields of the Northern Peruvian Jungle, as well as a more reliable Recovery Factor and EUR estimations at early development stages.