Assessing Uncertainty in Channelized Reservoirs Using Experimental Designs
- F. Friedmann (ChevronTexaco E&P Technology Co.) | A. Chawathe (ChevronTexaco E&P Technology Co.) | D.K. Larue (ChevronTexaco E&P Technology Co.)
- Document ID
- Society of Petroleum Engineers
- SPE Reservoir Evaluation & Engineering
- Publication Date
- August 2003
- Document Type
- Journal Paper
- 264 - 274
- 2003. Society of Petroleum Engineers
- 5.5.2 Core Analysis, 4.5.2 Platform Design, 5.6.9 Production Forecasting, 5.5 Reservoir Simulation, 1.6 Drilling Operations, 5.3.4 Reduction of Residual Oil Saturation, 5.4.1 Waterflooding, 3.3.6 Integrated Modeling, 2.4.3 Sand/Solids Control, 5.2.1 Phase Behavior and PVT Measurements, 1.2.3 Rock properties, 4.6 Natural Gas, 4.1.2 Separation and Treating, 6.1.5 Human Resources, Competence and Training, 4.1.5 Processing Equipment, 5.1.5 Geologic Modeling, 5.1.1 Exploration, Development, Structural Geology, 7.2.3 Decision-making Processes, 5.6.1 Open hole/cased hole log analysis, 5.7.2 Recovery Factors
- 1 in the last 30 days
- 870 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 12.00|
|SPE Non-Member Price:||USD 35.00|
It is well established that uncertainty exists in simulated recovery forecasts because of the ambiguity in the measurement and representation of the reservoir and geologic parameters. This is especially true for immature projects, such as deepwater reservoirs, where the high cost of data limits the information that is available to build reservoir models. We present two strategies based on Experimental Design (ED) to quantitatively assess this uncertainty in recovery predictions for primary and waterflood processes.
We apply the ED methodology to channelized sandstone systems because of their relevance to many deepwater projects. We choose to study synthetic geological analogs of channelized systems that are built from a panoply of relevant parameters while taking into account the uncertainty that exists in the estimation of their ranges. We use the results of this study to generate type curves with neural networks. The trained neural networks can be used to predict reservoir performance rapidly where field data are very limited. We discuss applications of this methodology on field cases from western Africa.
An irony of contemporary petroleum exploration and development is that although technological advances have been significant, risk has not been reduced in all cases. In fact, risk actually may be greater in many frontier reservoir development projects. For example, in deepwater projects, the large initial capital investments resulting from costs associated with platform design and well construction are made with limited knowledge of reservoir architecture and geology. The high cost of drilling, completing, and coring wells limits the availability of geological, petrophysical, and engineering data, which are needed to build reliable reservoir simulation models to help in the decision process. A method that can identify the key parameters governing uncertainty in production and economic forecasts in the early phases of the study will significantly ameliorate the data acquisition program.
Simulation is often the tool of choice in the planning and evaluation of sequential reservoir development phases. Typically, earth scientists build the most representative geological model, applying expert knowledge using well logs and other geological data. A few geostatistical realizations are generated to sample the uncertainty in geological parameters. A representative combination of geology, fluid, and flow parameters, along with well locations, constitutes the base-case model. This model is then simulated to obtain production profiles and recovery factor for a chosen recovery process. Finally, economic performance indicators [return on investment (ROI) and net present value (NPV)] are computed for the project.
Estimating recovery uncertainty is complicated because it requires an understanding of both the reservoir's static architecture and its dynamic behavior during production. Recovery depends on structural, stratigraphic, and per-meability architecture, fluid and engineering properties, drive mechanisms, and spacing/orientation of producing and injecting wells. The uncertainty associated with the measurement and estimation of these parameters will result in uncertainty in reservoir performance estimates.
Traditionally, the one-parameter-at-a-time approach is used to assess the uncertainty resulting from the various model parameters on the simulation results. Three levels per parameter (P10, P50, and P90) are defined. In this approach, one parameter is set at the P10 or P90 level, keeping all other parameters at the P50 level in each simulation. Tornado diagrams are then generated to rank each parameter effect on the recovery estimates. Probabilistic economics (e.g., NPV) are computed using a simple model based on the two or three most significant parameters in the tornado diagram. This method can be biased because its analysis relies on comparing the various simulations in which all but one parameter are maintained at the P50 level. Additionally, relatively limited information is obtained by running the (2N+1) simulations (with N representing the number of parameters) dictated by this technique. Sometimes, a very serious assumption is made whereby a P50 recovery case is identified when all the model parameters are maintained at their P50 levels. Additionally, it is implicitly assumed that all the variables being investigated are independent and do not interact with each other.
Another technique to assess uncertainty is to apply Monte Carlo methods directly to reservoir simulation. In this technique, the recovery uncertainty is estimated by assuming probability density functions (PDFs) for all the uncertain variables. Values for the uncertain variables are randomly sampled from these preassigned PDFs (which may be Gaussian, triangular, etc.) using a random number generator. The resulting distribution of the dependent variables, such as the recovery factor, is then evaluated by making repeated calls to the reservoir simulator. Unfortunately, this technique is very time-intensive; therefore, it is practically amenable to the simplest of reservoirs and has been restricted mostly to academic cases. One elegant way to overcome the time-intensive nature of this technique is to create a proxy model for the reservoir simulator. These proxy models are usually analytical representations of the reservoir simulator and can be evaluated very quickly. We use this technique to our advantage in this paper to assess uncertainty.
This work reports two strategies to quantify the uncertainty in simulation recovery forecasts. Both methods use the ED methodology extensively.1,2 The first method, based on recovery type curves, applies to highly immature projects in which the available data set is too sparse to build reliable geological models. The second method, based on response surface analysis, should be used for relatively mature projects, when sufficient data exist to construct geological models for flow-simulation studies. A summary of these strategies is given in Fig. 1.
Because of their relevance to many deepwater projects, we present the application of the ED strategies to assess uncertainty in the simulation of channelized sandstone reservoirs undergoing primary depletion and waterflood.
|File Size||3 MB||Number of Pages||11|