Sanjay Sirinivasan

(Penn State Department of Energy and Mineral Engineering)

Assimilation of dynamic data into reservoir models using a model selection approach

What GR HOMEPAGE Meteo Colloquium UG
When Sep 27, 2017
from 03:30 pm to 04:30 pm
Where 112 Walker Building
Contact Name Fuqing Zhang
Contact email
Add event to calendar vCal

Sanjay Srinivasan PSU EME

More information about Sanjay Sirinivasan

  • Head, John and Willie Leone Family Department of Energy and Mineral Engineering
  • Professor of Petroleum and Natural Gas Engineering
  • John and Willie Leone Family Chair in Energy and Mineral Engineering

Iterative adjustment of stochastic random fields (such as spatial variation of permeability in a reservoir) so as to match measured dynamic data can be treated as a classical, ill-posed inverse problem. A host of both heuristic and exact optimization techniques have been applied to problems with both discrete and continuous parameter spaces. Typically, these approaches yield model parameters that provide the best fit to the observed data and any remaining discrepancy between the estimated and observed values is regarded as uncorrelated noise.  The underlying theory of most optimization approaches is that this noise must follow the form of uncorrelated errors drawn from zero-mean, Gaussian distributions. In addition, the process of iteratively adjusting the large number of parameters (permeability nodes) renders the computation process expensive and time consuming. 

In lieu of this grid-based model updating process, we present a model selection algorithm that refines an initial suite of subsurface models representing the prior uncertainty to create a posterior set of subsurface models that reflect injection performance consistent with that observed. The process involves creating an initial suite of subsurface models that captures all the possible geological scenarios for the reservoir/aquifer under study; running all the models through a fast proxy; grouping models on the basis of the proxy responses; assigning a probability value to the most likely group of models by comparing flow simulation results for each group with the pressure history of injectors; and repeating this grouping method to iteratively create a posterior set of most probable models (a subset of the initial set). Such posterior models can be used to represent uncertainty in the future migration of fluids in the reservoir.

The model selection approach represents a clear departure from the iterative model perturbation approaches for dynamic data assimilation. It also offers the advantage of yielding a suite of models at the end of the data assimilation process. This permits assessment of residual uncertainty that enables making probabilistic statements regarding value of information etc. We demonstrate the method for tracking the migration of CO2 plume in the subsurface during sequestration. In that case, because only injection data is required, the method provides a very inexpensive method to map the migration of the plume and the associated uncertainty in migration paths. We illustrate the applicability of the method using a field data set.