Fractal-wavelet neural-network approach to characterization and upscaling of fractured reservoirs

Characterization of field-scale porous media (FSPM), such as oil, gas and g eothermal reservoirs, is a complex problem. Field data are often difficult to analyze because they exhibit complex patterns of behavior. The problem i s even more complex when one has to deal with fractured porous media. In th is paper we describe a new hybrid approach to comprehensive characterizatio n of FSPM, development of accurate fine-grid geological models for them, an d their upscaling. First, we describe the fractal approach to analyzing the porosity logs, the permeability distributions, and other properties of FSP M. We show that an accurate and efficient method of analyzing such data is provided by wavelet transformations. These transformations can also be used for the analysis of the patterns of fracture networks of FSPM, and process ing of seismic data, and therefore they provide a unified approach to the t reatment of practically all types of data for FSPM. We then propose a fract al neural network that can recognize fractal data and construct accurate co rrelations for estimating those properties of FSPM for which the data are s carce, e.g., their permeability distribution. Next, the results of the frac tal-wavelet neural-network model are combined with stochastic conditional s imulations to generate an accurate geological (fine-grid) model for FSPM, i ncluding their fracture network. A wavelet transformation method is then de scribed for upscaling of the geological model of FSPM. Thus, one has a unif ied approach to reservoir characterization and modeling in which three key concepts play prominent roles: fractal analysis, wavelet transformations, a nd fractal neural networks. (C) 2000 Elsevier Science Ltd. All rights reser ved.