A hybrid connectionist enhanced oil recovery model with real-time probabilistic risk assessment

Abstract

An effective enhanced oil recovery (EOR) method requires an evidence-based, data-driven assessment of the impacts of the governing parameters on the rate of oil production. In this research, an innovative and comprehensive approach is developed to analyze the optimization of oil production by carbon dioxide (CO2) injection and associated risks as a function of time in a shale formation reservoir. The data-driven approach is used to perform risk analysis. This study integrates a commercial simulator with a deep learning network and probabilistic method to predict oil production enhancement with the associated risks. The integrated approach presented in this study creates evidence from the Spatio-temporal model based on geological data of the shale formation. The data-driven model uses the long short-term memory (LSTM) network to forecast oil production based on Spatio-temporal datasets. The possible risks are assessed using a dynamic Bayesian network with an early warning index system (EWIS) derived from the data-driven model. The findings of the Spatio-temporal model show that CO2 injection would lead to the maximum oil recovery rate of 22% of the original oil-in-place with associated risks in each cycle of operation. This study confirms that the introduced model can serve as a multifunctional tool for oil production optimization with a clear understanding of associated risks.