Use of Multiple-Response Optimization To Assist Reservoir Simulation Probabilistic Forecasting and History Matching
- W. Terry Osterloh (Chevron Corp.)
- Document ID
- Society of Petroleum Engineers
- SPE Annual Technical Conference and Exhibition, 21-24 September, Denver, Colorado, USA
- Publication Date
- Document Type
- Conference Paper
- 2008. Society of Petroleum Engineers
- 5.6.9 Production Forecasting, 4.1.5 Processing Equipment, 5.1.5 Geologic Modeling, 5.6.3 Deterministic Methods, 5.4.6 Thermal Methods, 5.5.8 History Matching, 5.7.3 Deterministic Methods, 5.5 Reservoir Simulation, 4.1.2 Separation and Treating
- 2 in the last 30 days
- 420 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 8.50|
|SPE Non-Member Price:||USD 25.00|
This paper describes a straightforward, robust, and easily managed method for assisting reservoir simulation probabilistic forecasting, probabilistic history matching, and process optimization. The method, which is widely used in numerous other engineering fields, is based on design of experiment, Gaussian process modeling (kriging), and multiple response optimization. The method enables a new perspective on probabilistic green-field forecasting - creating a simulation model in which the probability numbers of multiple responses are at the same level (p-number alignment). The method also provides a probabilistic framework for history matching and forecasts made with the matched model. Example applications of p- number alignment and probabilistic history matching are presented for primary production and steamflood EOR pilots.
Highlights of the new method include the following: 1) multiple responses (even those with different units) can be aligned or matched, 2) responses can be weighted to focus the alignment or match on preferred responses, 3) ad-hoc, manual, or trialand- error adjustments of simulation inputs is largely eliminated, 4 ) the method can be used with any simulation code, and 5) the method is extensively automated, with all key steps integrated into one commercially-available statistics package.
Managing the risks associated with forecasts from numerical reservoir models has been a very active research topic within the petroleum industry. The risk inherent in using a single deterministic model forecast for making investment decisions in major capital projects in existing or new fields is widely recognized. The dominant risk is attributed to the substantial uncertainties in the subsurface model factors (inputs) which control the forecast responses (outputs). These factor uncertainties arise from sparse or often absent data; when factor data are available, uncertainty arises from random errors made in data collection and/or measurements, i.e. noisy data. The geological and engineering models are themselves imperfect representations of the subsurface geology and physics of fluid flow.
When historical production data is available, the numerical model factors are adjusted to calibrate the model to improve the match between the model prediction and the historical responses. This step is hoped to reduce the uncertainty in forecasting future production. However, this is a difficult, ill-conditioned problem which in the past produced a single deterministic, non-unique match that provided no information regarding forecast uncertainty. Much progress has been made in recent years to develop probabilistic frameworks around green-field (new fields with no history) forecasts and model calibration. Probabilistic model forecasts and calibrations provide assessments of project uncertainty and risk that were unavailable with the old deterministic methods. Additionally, researchers realized that these methods provided a new means
for assisting or substantially automating the model calibration process.
|File Size||244 KB||Number of Pages||10|