Applications of Data-Driven Techniques in Reservoir Simulation and Management

Advanced algorithms and supercomputers enhance the speed and accuracy of traditional modeling techniques (numerical and analytical); however, they are incapable of transforming the models to new levels of computational footprint. Application of artificial intelligence and machine learningMachine learning (AI&ML) in reservoir engineering has shown significant potential for transforming traditional modeling approaches. By leveraging data-driven techniquesData-driven techniques, AI&ML can develop powerful and fast tools that can model complex and uncertain physics. Such tools perceive the relationship among relevant data and develop models based on the available measurements or simulated data. Promising results have been obtained from the application of data-driven techniquesData-driven techniques for resolving a wide variety of reservoir managementReservoir management problems such as history matching, well control and placement, injection strategies optimization, production forecasting, CO2 storageCO storage, and many more. In this chapter (part 1), we aim to discuss some recent applications of ML-based data-driven models in reservoir simulationReservoir modeling and management, highlighting the benefits of such approaches in capturing high nonlinearity. We often refer to these data-driven models as proxy modelsProxy models since they act on behalf of the “actual” or physics-based models. The successfully trained proxy modelsProxy models are coupled with, for example, metaheuristic algorithms in solving optimization problems in reservoir managementReservoir management. We show examples of fast well control optimization in waterfloodingWaterflooding using proxy modelsProxy models including two-stage (local and global) proxy modeling to handle problems with high dimensions. Some relevant topics include integration of sampling techniques, adaptive sampling and retraining to address the geological uncertainties, and use of more complex reservoir models. To further illustrate the applications of machine learningMachine learning, we discuss: There is a need to develop advanced modeling techniques within the area of carbon capture, utilization, and storage (CCUS) for sustainable utilization of the subsurface. Such tools can quickly model and investigate the impact of uncertainties and decision variables on processes. In addition to the economic value of CCUS related to producing the remaining resources, improved modeling and optimization techniques will be useful when the subsurface is to be repurposed for storage. In this chapter (part 2), we aim to discuss further applications of data-driven models. We present examples where proxy modelsProxy models are employed in mono- and multi-objective optimization problems, for example in Water Alternating Gas (WAG) injection, in which design parameters are assessed to maximize the oil recovery and/or the CO2 stored. Development of proxy modelsProxy models are also discussed for simulation of CO2 injection and storage in Svelvik CO2 Field Lab in Norway, and Smeaheia CO2 storageCO storage in the North Sea. Accurate estimation of rock and fluid properties has a significant effect on the design and monitoring of CO2 storage. We will discuss data-driven techniquesData-driven techniques used to develop robust paradigms to accurately estimate the CO2 thermal conductivity and diffusivity in brine under various operating conditions using the representative experimental database. We also briefly discuss the use of machine learningMachine learning techniques in: Overall, the application of data-driven proxy modeling offers significant advancements in reservoir modelingReservoir modeling, addressing computational limitations and enabling efficient optimization and decision-making. Proxy modeling is however still subject to limitations including data samplingH storage efficiencyWater alternating gas, learning techniques, dimensionalityCO storage, uncertaintyGas injection, etc., thatReservoir management needWaterflooding furtherDecision analysis investigationsCarbon capture utilization and storage.