Full text loading...
-
Robust Optimisation Using Spectral High Dimensional Model Representation - An Application to CO2 Sequestration Strategy
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, ECMOR XIV - 14th European Conference on the Mathematics of Oil Recovery, Sep 2014, Volume 2014, p.1 - 9
Abstract
Successful CO2 sequestration relies on operation strategies that maximise performance criteria in the presence of uncertainties. Designing optimal injection strategies under geological uncertainty requires multiple simulation runs at different geological models, rendering it computationally expensive. A surrogate model has been successfully used in several studies to reduce the computational burden by approximating the input-output relationships of the simulator with a limited number of simulation runs. However, building the surrogate is a challenging problem since the cost of building the surrogate increases exponentially with dimension.
In the current work, we propose the use of Adaptive Sparse Grid Interpolation coupled with High Dimensional Model Representation (ASGI-HDMR) to build a surrogate of high-dimensional problems. This surrogate is then used to assist with finding robust CO2 injection strategies. High Dimensional Model Representation (HDMR) is an ANOVA like technique, which is based on the fact that high-order interactions amongst the input variables may not necessarily have an impact on the output variable; the combination of low-order correlations of the input variables can represent the model in high-dimensional problem. Adaptive Sparse Grid Interpolation (ASGI) is a novel surrogate technique that allows automatic refinement in the dimension where added resolution is needed (dimensional adaptivity).
The proposed technique is evaluated on several benchmark functions and on the PUNQ-S3 reservoir model that is based on a real field. For the PUNQ-S3 model, robust CO2 injection strategies were estimated efficiently using the combined ASGI-HDMR technique. Based on our numerical results, ASGI-HDMR is a promising approach since it requires significantly fewer forward runs in building an accurate surrogate model for high-dimensional problems in comparison to ASGI without coupling with HDMR. Hence, the ASGI-HDMR enables efficient construction of the surrogates for high-dimensional problems with several wells and over multiple control periods. The impact of complex and high-dimensional control strategies to the performance criteria is shown in our finding.