Reservoir data assimilation for realistic geology

Reliable reservoir simulation models are crucial for improved oil recovery at existing fields. The simulation models allow prediction of future reservoir behavior. This prediction forms the basis for decisions in reservoir management. For example, the models can be used to compute the optimal location of infill wells, or to compute the optimal management strategy for a reservoir. In this project, RESERVOIR DATA ASSIMILATION FOR REALISTIC GEOLOGY, ensemble based methods, such as Ensemble Kalman Filter (EnKF), are used to improve the reservoir models. (To simplify EnKF is used for ensemble based methods in the rest of the text.) In EnKF an ensemble of reservoir models is used, typically not more than 100 models to limit the computational time. All the models are updated accounting for information about the reservoir. The data used for updating are typically production data, data from well logs and sometimes seismic data. The EnKF is a Monte Carlo method where statistics calculated from the ensemble is used to update the models. The result is a new set of models giving an improved history match of the data and models that are better suited to predicting the reservoir behavior. An important benefit of using this method is that the ensemble of models provide an uncertainty estimate, both in the estimated models and when predicting the future. A robust uncertainty estimate is of crucial importance in all decision makings regarding petroleum production. One aim of this project is to include this robust and realistic geological uncertainty measure in the estimated ensemble of reservoir models. The fact that the EnKF is developed for weakly nonlinear problems and approximate Gaussian distribution of variables and measurement errors can be a challenge when e.g. dealing with realistic geology. Another challenge is that using 100 ensemble members might not be enough, and can lead to erroneous updates. In this project we have focused on developing the EnKF to characterize complex geological properties contained within real reservoirs (e.g. sudden jumps between facies types with different characteristics, faults, etc.) and to history match the models without losing the characteristic properties within the initial models.

Reliable reservoir simulation models are crucial for improved oil recovery at existing fields. The simulation models allow prediction of future reservoir behavior. This prediction forms the basis for decisions in reservoir management. For example, the mod els can be used to compute the optimal location of infill wells, or to compute the optimal management strategy for a reservoir. Due to the complexity of the reservoir, realistic geological models often include complex spatial dependencies of petrophysic al properties, discontinuities in petrophysical properties at facies boundaries, and inhomogeneous faulting and fracture system. Although a reservoir simulation model is often an up-scaled version of the geological model, a large number of model parameter s is still necessary to appropriately simulate the actual reservoir behavior. Because of the high dimensionality of the model and the non-Gaussian prior probability distributions of the model parameters due to complex geology, obtaining calibrated simulat ion models with plausible geological characteristics and reasonable match to dynamic data is exceptionally difficult. Most history-matching algorithms can only handle a limited number of parameters. This reduction in the number of degrees of freedom in hi story matching often results in models that are not geologically plausible, give poor matches to well data and consequently cannot provide reliable predictions. Some recent ensemble-based history matching methods have the ability to adjust large numbers o f parameters. The standard use of these methods, however, results in reservoir models that match data but are not geological plausible, so that the confidence in future predictions is reduced. The primary objective of this project is to develop methodolo gies for characterization of the complex geological features of the reservoir and to history match models in a geologically consistent manner for improved predictability and quantification of uncertainty.