The use of flow diagnostics to rank model ensembles

Abstract

Ensembles of geomodels provide an opportunity to investigate a range of parameters and possible operational outcomes for a reservoir. Full-featured dynamic modelling of all ensemble members is often computationally unfeasible, however some form of modelling, allowing us to discriminate between ensemble members based on their flow characteristics, is required. Flow diagnostics (based on a single-phase, steady-state simulation) can provide tools for analysing flow patterns in reservoir models but can be calculated in a much shorter time than a full-physics simulation. Heterogeneity measures derived from flow diagnostics can be used as proxies for oil recovery. More advanced flow diagnostic techniques can also be used to estimate recovery. With these tools we can rank ensemble members and choose a subset of models, representing a range of possible outcomes, which can then be simulated further. We demonstrate two types of flow diagnostics. The first are based on volume-averaged travel times, calculated on a cell by cell basis from a given flow field. The second use residence time distributions, which take longer to calculate but are more accurate and allow for direct estimation of recovery volumes. Additionally we have developed new metrics which work better for situations where we have a non-uniform initial saturation, e.g., a reservoir with an oil cap. Three different ensembles are analysed: Egg, Norne, and Brugge. Very good correlation, in terms of model ranking and recovery estimates, is found between flow diagnostics and full simulations for all three ensembles using both the cell-averaged and residence time based diagnostics.

Rapid Flow Diagnostics for Prototyping of Reservoir Concepts and Models for Subsurface Co2 Storage

“…To obtain further dynamic properties for FD that approximate the reservoir behaviour such as well allocation factors, drained or swept reservoir regions, influence regions that are affected by the interaction between wells, or the volume of fluid displaced by the injected fluid as a function of time, it is convenient to compute the steady-state concentration c of an inert and virtual tracer for the reservoir (Lie et al, 2015;Møyner et al, 2015;Rasmussen and Lie, 2014;Watson et al, 2021) v⋅∇c = 0, c| inflow = 1 (4) Shook and Mitchell (2009) demonstrated how other characteristics that approximate the dynamic reservoir behaviour, such as the dynamic Lorenz coefficient, storage efficiency, or breakthrough times, can be calculated from the distribution of τ in the reservoir. Using the storage capacity Φ(τ) and flow capacity F(τ) given by Møyner et al ( 2015)…”

“…these reservoir models in an intuitive and time-effective way (Alshakri et al, 2023;Jackson et al, 2022). The resulting model ensemble can be formally compared using clustering techniques based on specific reservoir performance metrics (Scheidt and Caers, 2009;Watson et al, 2021).…”

“…First, new boundary flow conditions are introduced to enable calculation of the storage potential for injected CO 2 . Previous FD works considered wells as boundary conditions and assumed all outer boundaries to be closed (Watson et al, 2021;Zhang et al, 2020). Second, grid resampling 'on-the-fly' is integrated with FD to discretise the surface-based representation of the reservoir geology, which allows grids to be updated while assessing the impact of geological heterogeneity on flow behaviour using FD.…”

Poro-Mechanical Coupling for Flow Diagnostics

“…For this particular example, the entire computations were carried out on a standard desktop PC for the Tarbert and Ness formations; it took only 16 and 23 min to simulate the whole workflow, respectively. Such fast computations allow us to analyse a much wider range of geological scenarios, parameter combinations, and well patterns, and hence enable us to screen and explore a broader range of uncertainties prior to selecting models and scenarios for more detailed full-physics simulations using appropriate clustering and ranking techniques (Caers and Scheidt 2011;Scheidt and Caers 2009;Park et al 2013;Watson et al 2021). Performing similar screening simulations would likely take days using coupled full-physics simulations.…”

“…Results from flow diagnostics can therefore be used to select a small subset of reservoir models based on their dynamic performance from a large model ensemble without compromising on the ability to forecast future reservoir performance and estimate the impact of uncertainties (e.g. Caers and Scheidt 2011;Scheidt and Caers 2009;Park et al 2013;Watson et al 2021). Our proposed poro-mechanical integration with flow diagnostics frameworks hence provides us with a computationally efficient estimation of how poro-mechanics might impact reservoir performance, which allows us to compare, contrast, and rank different stress-sensitive reservoir models and select individual scenarios for further detailed studies that use computationally demanding full-physics models that couple geomechanical and hydrodynamical simulations.…”