With heterogeneous reservoirs, ther eis a high risk of drilling wells in no
nreservoir facies. An evaluation of this "dry well" risk is necessary for d
ecision making. The "dry well" risk probability is defined as a nonlinear f
unction of thestochastic reservoir model parameters (facies proportions and
their correlation lengths) and the well parameters (production interval an
d orientation). There exists an analytical formula of the "dry well" risk p
robability for a class of stochastic reservoir models which have the semi-M
arkovian property (e.g. the Boolean model with convex elementary objects).
This paper proposes a generalization of this formula to other stochastic re
servoir models commonly used in practice.
The proposed approach is validated on two numerical reservoir models genera
ted respectively by the most commonly used geostatistical methods: the sequ
ential indicator simulation and the truncated Gaussian simulation. The cons
truction of these reservoir models is inspired from a real field. In both c
ases satisfactory approximations of the "dry well" risk probability are obt
ained Unlike a conventional experimental design based on a polynomial appro
ximation, the proposed approach gives a geological signification to the par
ameters which influence the "dry well" risk probability.
The method can be used to optimize the orientation and the completion (prod
uction interval) of a new well. In particular, it can be used to evaluate t
he efficiency of a horizontal well as opposed to a vertical or inclined wel
l. When evaluating the uncertainty on the performance of a new well, it is
convenient to evaluate first the "dry, well" risk and then to evaluate the
well performance provided that the well hits the reservoir facies.