Abstract
We tackle reservoir simulation model initialization in situations where the current free water level has risen significantly above the original paleo-contact. In such cases, traditional initialization based on primary drainage capillary-gravity equilibrium is insufficient because it does not consider the reimbibition that takes place after primary oil migration. Usage of traditional initialization would require a generation of several thousands of regions to have a satisfactory representation of such phenomenon in the model.
We apply this initialization to a reservoir with a complex initial fluid in which hydrocarbons come from multiple sources resulting in significant spatial variation. To capture this spatial variation, the scripting feature of a commercial simulator is used to generate on-the-fly PVT regions and fluid descriptions based on continuous fluid properties. Capillary-gravity equilibrium is then used to initialize the model with the primary drainage capillary pressure curves at the paleo-contact, which gives the minimum water saturation for each grid-block. From this, an imbibition scanning curve is generated for a second initialization based on the current oil-water contact thereby including reimbibition.
Our proposed approach was applied to a giant carbonate field for which we were able to generate on-the-fly PVT regions based on continuous property maps of fluid properties such as oil API, solution gas-oil-ratio and water salinity. This resulted in fewer regions being generated and thereby reducing the amount of memory required to initialize the model. We were also able to reduce the initialization time as compared to use of discretized contacts and regions. Our approach enabled us to initialize with both the paleo- and current contacts allowed to continuously vary spatially. This together with the use of a more appropriate capillary pressure hysteresis model, resulted in an initial reservoir state that gave a better match to the saturations obtained from logs.
This work illustrates the creation of PVT data and fluid regions on-the-fly based on continuous fluid properties. We also demonstrate the use of continuously varying paleo- and current oil-water contacts without the need to discretize these into regions. These aspects facilitate the propagation of an uncertainty workflow, starting with continuous fluid properties and structural modelling, directly to the simulation model with no need for intermediate discretization and without massive computational penalties.
