Full-Physics, Streamline-Based Method for History Matching Performance Data of a North Sea Field

Abstract

Abstract Streamline-based methods for history matching are appealing for two reasons. First, the forward simulation is potentially much faster than conventional simulation methods for displacement-type problems. Second, time-of-flight information along the streamlines can be used to find sensitivity coefficients in an efficient and elegant manner. However, current streamline history matching methods use forward simulations that ignore the effects of gravity and compressible flow. These effects may be critical in analyzing early-time performance data. If relevant physical effects are neglected in the forward simulation, the history matched reservoir description may not be consistent with the actual field, and predictions from the model will not be reliable. We present a method for history matching watercut data using a streamline simulation that captures all the pertinent physics, including compressible three-phase flow with gravity. We use a methodology based on the assumption of one-dimensional flow along streamlines to find the sensitivity of water flow rate at production wells to changes in permeability. Although the computation of the sensitivities is approximate, the method provides a good history match for problems with significant effects due to compressibility and gravity. Data from a North Sea field is used to test the technique. Using a full-physics streamline model gives a reasonable history match and a good prediction of future performance. Introduction Several history matching methods have been proposed to constrain reservoir descriptions to production data, such as water-cut history (see, for instance1–8). These techniques use conventional grid-based simulation to compute sensitivity coefficients, which give the change in production data due to a change in the permeability or porosity of some portion of the simulation model. Using the sensitivity coefficients, the porosity and permeability data are adjusted to create a new reservoir model. When another simulation is performed using this model, a better match to the data should be obtained. If the match is still not acceptable, new sensitivity coefficients are computed and used to modify the reservoir model again. Because the sensitivity coefficients are non-linearly dependent on the reservoir description, many iterations may be needed before a good history match is obtained. For a finely-gridded model, there are many more matching parameters than data, and the match is non-unique. To overcome the problem of having many poorly determined parameters in the reservoir description, methods that reduce the parameter space, such as the use of recursively refined grids,9,10 gradzone analysis2,4 and gradual deformation7,8 have been proposed. Current history matching methods are still limited, however, by the time taken to perform the forward simulations. Ideally, history matching would start from a statistical ensemble of equi-probable initial estimates of the reservoir description. Each reservoir description, if it is a fine-scale reservoir model, may contain several million grid blocks. Current simulation techniques may take approximately a week to perform a single simulation of this size, making history matching, that may require tens to hundreds of iterations, on several models, impossible in practice. Approaches where history matching is performed on a much coarser grid than the geological model offer appealing savings in computer time. However, the appropriate manner to upscale or downscale single and multiphase flow properties is not obvious. Furthermore, fine-scale details may be important for future prediction of, say, waterflood or gasflood performance. In these cases, history matching on a coarse grid may yield a poor reservoir description with correspondingly inaccurate predictions of future recovery.