An integrated framework based on deep learning algorithm for optimizing thermochemical production in heavy oil reservoirs

Abstract

The development of heavy oil reservoirs with active edge and bottom water is one of the most challenging problems in petroleum engineering. In response to the limited thermal recovery of these reservoirs, a multi-phase and multi-component numerical simulation model for thermal and chemical recovery is proposed. An edge-water assisted chemical flooding (EAC flooding) is proposed, which can improve oil displacement efficiency and sweep efficiency by rational utilization of edge-water energy when compounding multi-component chemical system. Then, a deep reinforcement learning algorithm is proposed to predict dynamic production parameters and determine the optimal working system to maximize the oil recovery according to the above mathematical model. The deep reinforcement learning (DRL) model can predict the dynamic production curves according to given states with optimal strategy. At the same time, the proposed model can determine the best conversion timing from cyclic steam stimulation to EAC flooding. Finally, the DRL model can automatically obtain the optimal working system, effectively improving the oil recovery while considering the economic benefits. Thus, the DRL model can solve traditional numerical simulation's time-consuming and labor-intensive challenges and accurately give the optimal working system for developing heavy oil reservoirs with edge water in the field.