• Home
• Program
• Search
• Author Index
• Sponsors
• Committees
• Contact Us
• Tech Support

Session:

Petroleum 3

Abstract No.:

O-043

Title:

Uncertainty Quantification For History-Matching Of Non-Stationary Models Using Geostatistical Algorithms

Author(s):

Maria Helena Caeiro, Instituto Superior Técnico (PT)
V. Demyanov, Heriot-Watt University (UK)
A. Soares, Instituto Superior Técnico (PT)
M. Christie, Heriot-Watt University (UK)

Abstract:

Uncertainty quantification associated to the characterization of a petroleum reservoir is a challenge in reservoir modelling practices. Describe the fluid flow course of meandering channels is a prime issue in the history matching of non-stationary processes. A novel method to integrate image transforming method based on direct sequential simulation and co-simulation with local anisotropy correction (DSS-LA) and sampling optimisation techniques in the history matching process is presented. DSS-LA is a variogram based algorithm which has the advantage of using the original variable without requiring any processing, very relevant for the simulation of continuous variables. It tackles the problem of non- stationarity and connectivity of the model by introducing spatial trends, which represent local anisotropy variations (direction of maximum continuity and anisotropy ratio). The parameters of the local anisotropy model are optimised to match the production history data, which leads to uncertainty quantification through the Bayesian inference. The methodology was applied to a synthetic petroleum reservoir with fluvial deltaic structures (Stanford VI). As a result, the stochastic models of reservoir?s properties (porosity and permeability) and the respective dynamic responses are obtained. Multiple history matched models quantify uncertainty related to the trend model. They have a good fitting to the production data and provide more precise uncertainty evaluation for the predictions. This application aims to demonstrate the feasibility of linking DSS-LA with an optimisation algorithm to integrate dynamic data and quantify the uncertainty associated to the characterization of a non-stationary reservoir in the history matching process. The results reveal that this has been successfully achieved and the proposed approach demonstrates efficiency for the generation of multiple history matching models.