Stochastic Modeling of Fluvial Reservoir: The YACS Approach

Abstract

Abstract This paper describes a novel approach to model fluvial reservoirs. The core of the method lies in the association of a fairway or river bed with the channels to be simulated. A potential field is defined within the fairway. Specifying a transfer function to map this potential field to thickness results in a generating a channel inside the fairway. A residual component is stochastically simulated and added to the potential field creating sinuosity and realistic channel geometries.The degree of deformation is controlled and allows the creation of channels with high sinuosity including oxbows. Conditioning to well data is obtained by applying the inverse transfer function at the data location to derive thickness values which will constrain the simulation of residuals. Fairways are positioned so that all data are taken in account; they can be stochastic if unknown or explicitly entered if identified on seismic data. The method allows simulating channels which honor both exactly hard well data and channel proportions, following arbitrary shapes guided by fairways. The algorithm is based on sequential algorithms which guarantee convergence and speed; there is no iteration or optimization loop. This new approach can simulate channel, levees, and also lobes; channels can have multiple branches and conditioning a single channel (branch) to multiple wells is also possible. Introduction Strochastic modeling of fluvial-deltaic reservoirs generally requires specifically tailored algorithms which aim at representing realistic geological objects typically found in these environements: these include channels, associated levees and crevasses, and deltaic or turbiditic lobes. These reservoirs have the pecularity of exihiting a clear contrast of very high porosity and permeability rocks in the channel sands and very low porosity and permeability rocks in the mud of the flood plain. These contrasts determine where the hydrocarbons are and how fluid will flow through the reservoir. A reservoir model needs to accurately position channel sands in order to understand the reservoir behavior (past and future) and appropriately plan field development; this is especially true in lower net-to-gross reservoirs where clear channel shapes are present. Connectivity of the channels and their tortuosity determine fluid flow and ultimately production. To construct an accurate reservoir model, it is therefore necessary to properly capture the inter-connectivity of the channel system and the possible meandering of individual channels, i.e. generating geologically realistic features. There are two principal stochastic simulation approaches commonly used to model fluvial reservoirs: object-based algorithms (Georgsen and Omre [1993], Deutsch and Wang [1996], Holden et al. [1998]) have been traditionnaly applied but more recently multiple-point statistics (MPS) techniques have been introduced (Strebelle [2002]). The issue at hand is modeling long sinuous objects (the channels) and placing them in the reservoir grid such that all well data is honored and any local proportion constraint is respected. These methods will not be discussed in detail here, see references. Simplifying:object-based techniques use iterative optimization algorithms to place objects with predetermined shapes at their most likely location leading to two main short-comingsit is difficult and extremely time consuming to honor all constraints when there are many wells (~>30)resulting models do not look geologically realistic and it is often not possible to model highly meandering channels;MPS require a "training image" which is a 3D "realistic" representation of the reservoir's geology but not constrained to any well data, it then reproduces observed patterns honoring perfectly well information and fairly well any other secondary proportion constaints, howevercontinuity of long objects can be a problem,the patterns are reproduced as a whole and there is no knowledge of individual objects. Alternatively, one could look at process-based methods to generate channels (Bridge and Leeder [1979], Lopez [2003]); these have the advantage to generate realistic images of the geology however it is difficult to constrain them to well data and they are time-consuming.