History Matching Under Geological Control: Application to a North Sea Reservoir

Abstract

Solutions to inverse problems are required in many Earth Science applications. The problem of determining reservoir properties, such as porosity and permeability from flow data, shortly termed ``history matching'', is one example. In many traditional inverse approaches, certain model assumptions are made on either the data likelihood or the prior geological model, e.g. assumptions of conditional independence between data or Gaussianity on the distributions, which do not reflect the reality of actual data. This limits the applications of such approaches to practical problems like history matching. While modeling assumption are inevitable, this paper presents a general inversion technique that can be used with different geostatistical algorithms to create models that honor several types of prior geological information and at the same time match almost any type the data. The technique is built on the idea of perturbing the probability distributions used to create the models rather than perturb the properties directly. By perturbing the probabilities, the prior geological model as described by a geostatistical model or algorithm is maintained. We present a practical implementation of the probability perturbation method. A case study demonstrates how the practical implementation would work in an actual situation. The case study is a North Sea hydrocarbon reservoir where the production rates and pressure information are iteratively included in the model.