Integrated Reservoir Description, SPE 24261-MS

The main challenge in reservoir description is to integrate the different data sources and in particular be able to handle the uncertainty in the reservoir description. This paper presents a stochastic approach which integrates the different data sources. The method presented may be used for improving the predictions of the performance and for assessing the associated uncertainties. The latter is, however, a more complicated task.

The method consists of a combination of many different techniques, each of them used to model particular reservoir phenomena. The examples shown includes the use of Gaussian random fields, marked point processes and Markov random fields and the respond variables are hydrocarbon in place, oil production and recovery factor.

Introduction

There are many different data sources with relevant information, e.g. well data, seismic data, well tests, production history, outcrops and general geological knowledge. These data sources have very different properties. The main challenge in reservoir evaluation is to integrate these different data sources consistently and in particular quantify the large uncertainty which still is left. In this paper a stochastic approach is described. Stochastic models constitute the only method for formal quantification of uncertainty. In addition stochastic models provide the opportunity to model the different precision in the data sources and model general properties like smoothness of surfaces and correlation between variables. It is important to be able to update the reservoir description efficiently. Stochastic models also seems wellsuited to handle this problem. Haldorsen and Damsietti, (1) give a similar list of arguments in favor of stochastic models.

In this paper we will concentrate on how to integrate the different data sources. Only the aspect most important for petroleum recovery should be modeled. For each aspect a range of different models are available. Since the main issue in this paper is integration, we will only use examples which the authors have worked with personally. References are given to some of the other methods. For an overview see [1], [2] and (3].

The approach presented in this paper consists of a combination of models in order to describe the different aspects of a reservoir. The approach will be demonstrated with several examples from different case studies on realistic data. The different modules of this stochastic approach is implemented and tested out, but it is not implemented as one system.

The authors believe that it is important to model the reservoir with as fine grid as possible. With today's technology this is typically 10 - 10 blocks. A fine grid is important for verification of the model. The most important verification is done by visual inspection of the realizations by experienced geophysisists and petroleum geologists. Different statistical indications, e.g. compare statistics from fake well with statistics from observed wells are also useful. Formalized models ensures that the realizations satisfies certain specified properties, e.g. stationarity and intercorrelation. This may not be ensured if ad hoc algorithms are used. In formalized models it is possible to estimate parameters formally and estimate the uncertainty in the predictions. Also in the prediction of the fluid flow it is important to use as fine grid as possible. With today's technology this is 10 - 10 blocks.