A New Methodology for Simulation and Optimization of CO2 Sequestration in a Saline Aquifer Using Artificial Neural Network and Model Predictive Control Approach

Carbon dioxide sequestration efforts are the subject of intense research due to growing international concerns over CO2 emissions critically affecting the global climate change. Extensive studies are being done to determine the extent of the CO2 injection so that CO2 storage is secure and environmentally acceptable. In this work, CO2 injection into a saline aquifer is used to simulate an aquifer in Eclipse-100 using a black oil reservoir simulator whose consequent behavior is predicted by an artificial neural network, being trained via the Levenberg-Marquardt algorithm. A generalized predictive control scheme is accommodated in the proposed design methodology to determine the optimal injection rates during the CO2 injection process, leading to an ultimate goal of maximized residual trapping or equivalently minimized structural trapping of carbon dioxide. This will eventually maximize the total volume of the injected CO2, while diminishing the risk of its leakage to the atmosphere due to a fail in the integrity of the formation cap rock. A set of diverse test scenarios has been organized to demonstrate the capabilities of the proposed methodology to tackle the probable practical considerations.