Top Down Intelligent Reservoir Modeling

Abstract

Abstract This paper examines the validity of a recently introduced reservoir simulation and modeling technique. The technique, that is named Top-Down Intelligent Reservoir Modeling, TDIRM (not to be confused with BP's TDRM history matching technique), integrates traditional reservoir engineering analysis with Artificial Intelligence & Data Mining (AI&DM) technology in order to arrive at a full field model and to predict reservoir performance in order to recommend field development strategies. The distinguishing feature of this technology is its data requirement for its analysis. Although it can incorporate almost any type and amount of data that is available in the modeling process, it only requires field production rate and some well log data (porosity, thickness and initial water saturation) in order to start the analysis and provide a full field model. Presence and incorporation of other types of data can increase the accuracy and validity of the developed model. In this work three different reservoir models with varying criteria have been generated using a commercial simulator. The models are built to investigate TDIRM's capabilities in predicting different aspects of an oil reservoir. The models include different reservoir saturation conditions (saturated or under-saturated), different number of wells and different distributions of reservoir characteristics. Production rates and well log data from the wells in the simulation model are imported into the TDIRM to develop a new empirical reservoir model and make predictions on new well performance and potential infill locations. The results from the TDIRM analysis are compared to the original simulation models. Investigation and validation of TDIRM's predictive capabilities include identification of gas cap development in the formation, identification of infill locations by mapping the remaining reserves as well as predicting flow performance of the wells that are planned to be drilled in the reservoir. Introduction Understanding the reservoir depletion and achieving high recovery factor has always been a challenge for reservoir engineers. In the past several years efficient techniques have been developed to study reservoir behavior and to build models that would allow analysis and predictions, however, the techniques that are based on numerical solution of the fluid flow equation required a large amount of reservoir data and are expensive from a time and man-power stand point. The analytical approaches to building such models usually limit the analysis to single-well models and include approximations and assumptions that limit their use for full field analysis. In recent years, a new empirical modeling technique has been introduced that is called Top-Down Intelligent Reservoir Modeling (TDIRM) (1) (2) (3) that approaches full field reservoir modeling from a different angle. The TDIRM's advantage is its flexibility in data requirement. It needs only production rate data and well logs (for some wells not all) in order to start its analysis and build full field model. It has also the capability of incorporating other data such as core analysis, well tests, pressure data and seismic, in cases where such data is available. The main disadvantage of TDIRM is that it is recommended to be used in fields with at least 50 wells and about five years of production history. This has to do with the fact that TDIRM uses the production history and well log data in order to generate a large spatio-temporal database of the reservoir static and dynamic behavior. It uses Artificial Intelligence and Data Mining (4) (5) (6) techniques to deduces field-wide patterns from the large spatio-temporal database. The result of these analyses is a full field model with impressive predictive capabilities.