Page 178 - Principles of Applied Reservoir Simulation 2E
P. 178

Part II: Reservoir Simulation  163


       that are nearest neighbors to the central block along orthogonal Cartesian axes.
       In Table 16-1, the central block is denoted by "C" and the nearest neighbor block
       contributing to the standard  finite difference  calculation in 2D are denoted by
       an asterisk. The five blocks comprise the five-point differencing  scheme of the
       2D Cartesian grid.
                                   Table 16-1
                            Finite Difference  Stencils
                    Block       1-1          I         1 + 1

                    J-l          9           #           9
                      J           *          C           *

                    J + l        9           *           9


             Reservoir  simulators  are usually formulated  with the  assumption  that
       diagonal blocks do not contribute because the grid is aligned along the principal
       axes of the permeability tensor. Diagonal blocks are denoted  by "9" in Table
        16-1. The nine-point stencil includes  all nine blocks in the calculation  of flow
       into and out of the central block. Grid orientation effects  can be minimized, at
       least  in principle,  if the  diagonal blocks are included  in the  nine-point  finite
       difference formulation [for example, see Young, 1984; Hegre, etal, 1986: Lee,
       et  al.,  1997]. This  option is available  in some  commercial  simulators. In 3D
       models, the number of blocks needed to represent all adjacent blocks, including
       diagonal terms, is 27. By contrast, only seven blocks are used in the conventional
       formulation  of a 3D finite difference  model.
             Local grid refinement  (LGR)
                                             LGR             Radial  Grid
       is  used  to  provide  additional  grid
       definition  in a few selected  regions
       of a larger grid. Raleigh [1991] com-
       pared  local  grid  refinement with a
       radial grid (Figure 16-8) and showed
       that  the  results  are  comparable.
       When LGR   is used, it typically in- Figure  16-8. LGR and radial  grids,
       creases  computer processor  time for a run because of increased  throughput in
   173   174   175   176   177   178   179   180   181   182   183