Empirical Aids to Reservoir Fluid Simulation: Using Data Driven Models to Augment Compositional Simulation

Abstract

Reservoir simulation is one of the most critically important tools in the modern reservoir engineering arsenal. Its uses range from reserves estimation to fluid and pressure dynamic simulation to well planning, field development and production forecasting. Reservoir fluid models are a major part of these reservoir models and are critical to the performance of the reservoir model. Traditionally, reservoir fluid simulation has been pursued using two general approaches: i) Black oil simulation and ii) Compositional fluid simulation. While black oil simulations are much faster in terms of computational overhead, they are not as accurate and may fail outright in case of near-critical crude oil mixtures. On the other hand, compositional fluid models are much more accurate but require more time and computational power to solve. The industry norm is now shifting towards compositional simulation due to easy availability of computing power and requisite gain in accuracy. However, there is a ‘time lag’ associated with the deployment of these models. Compositional models use an Equation of State (EoS) for computing the fluid densities and volume behaviour. These models need to be calibrated or ‘tuned’ before they are applied to the simulation model. Traditionally, the time required for acquisition and experimentation before data release for tuning takes over a year. This may prove to be a disadvantage in frontier exploration areas, or in complex compositionally graded reservoirs. This paper presents a new method for increasing the accuracy of non-tuned Equation of State (EoS) models that is referred henceforth as augmentation. This augmentation process utilises data driven models to estimate reservoir fluid parameters such as Bubble Point Pressure, Oil Formation Volume Factor or Gas Oil Ratio. These estimated/derived parameters, if derived from data from local data sets with geological or analogical similarities to current reservoir/basin under consideration are more accurate than the ‘un-tuned’ model. This gain in accuracy is transferred from the data driven models to the compositional models through the proposed augmentation technique before it is actually tuned to the actual fluid data. Improvements in accuracy up to 40% were observed on using this technique as opposed to standard non-tuned simulations. This technique facilitates the development of more accurate simulation models at an earlier stage of the reservoir development workflow than traditionally methods.