Data Driven History Matching for Reservoir Production Forecasting
Abstract/Contents
- Abstract
- Prior to performing reservoir forecasting, an inverse problem is often solved, in which one or more models are inferred from historical data and prior geological information. For large models, this procedure can be very expensive computationally, and it can also be challenging to maintain geological realism in the resulting model. These issues motivate the development of alternative approaches for reservoir forecasting. In this thesis we develop one such procedure, which we refer to as data driven history matching or DDHM. In our DDHM method, multiple geostatistical reservoir models, consistent with prior geological information, are first created and simulated to provide prior simulation data. Reservoir forecasts are then obtained by linearly combining the prior simulated data vectors with associated weights. The weights are computed through linear regression on historical data. The maximum a posteriori (MAP) estimate is computed with DDHM by performing singular value decomposition on an appropriate data matrix. This procedure is applied for production forecasting for a three-dimensional Gaussian model. The DDHM forecast shows clear improvement compared with the forecast from the `best` prior model. However, unphysical results (e.g., negative rates) are observed in some cases, presumably due to the existence of nonlinear features in the prior simulated data that are not modeled in the basic DDHM procedure. To address this issue, we introduce a series of mapping operations to transform the prior simulated data to data that are closer to Gaussian and more nearly linearly correlated. This is shown to improve DDHM predictions and to lead to MAP estimates that are physically reasonable and accurate, for both Gaussian and channelized geological models. We next incorporate the DDHM algorithm into a randomized maximum likelihood (RML) procedure for uncertainty quantification. This RML-DDHM method is tested on Gaussian and channelized models. A computationally intensive rejection sampling algorithm is used to provide reference P10-P50-P90 estimates of posterior uncertainty. Reasonably close agreement between RML-DDHM and rejection sampling is observed in both cases, which suggests that the RML-DDHM procedure is able to provide useful uncertainty assessments. Finally, RML-DDHM is applied to more complicated cases involving up to 16 wells. For the cases considered, the range of forecast uncertainty is reduced significantly relative to the prior uncertainty, and the true data consistently fall within or very near the RML-DDHM results. Taken in total, our findings suggest that a data driven approach such as DDHM may eventually represent a viable alternative to time-consuming model-based inversion procedures.
Description
| Type of resource | text |
|---|---|
| Date created | August 2014 |
Creators/Contributors
| Author | Sun, Wenyue |
|---|---|
| Primary advisor | Durlofsky, Louis J. |
| Degree granting institution | Stanford University, Department of Energy Resources Engineering |
Subjects
| Subject | School of Earth Energy & Environmental Sciences |
|---|---|
| Genre | Thesis |
Bibliographic information
Access conditions
- Use and reproduction
- User agrees that, where applicable, content will not be used to identify or to otherwise infringe the privacy or confidentiality rights of individuals. Content distributed via the Stanford Digital Repository may be subject to additional license and use restrictions applied by the depositor.
Preferred citation
- Preferred Citation
- Sun, Wenyue. (2014). Data Driven History Matching for Reservoir Production Forecasting. Stanford Digital Repository. Available at: https://purl.stanford.edu/yj791yc8536
Collection
Master's Theses, Doerr School of Sustainability
View other items in this collection in SearchWorksContact information
- Contact
- brannerlibrary@stanford.edu
Also listed in
Loading usage metrics...