The Pennsylvania State University
Formulating models that are data-exact, reproduce target statistics and at the same time provide an assessment of uncertainty due to incomplete information is an important objective of stochastic spatial modeling. Ultimately, these models for uncertainty feed into models for decision-making and reservoir development.
In this talk, I will present some of my recent efforts to stochastically model complex reservoirs, develop fast transfer function proxies to represent flow and transport through such complex reservoirs and finally, perform real-time updating and feed-back control of reservoir processes using these fast proxies. I will conclude my talk with some thoughts on how these improved uncertainty quantification procedures can help critical questions such as the value of information and the timing of reservoir develop decisions.
Geological systems such as subsurface reservoir or aquifers in many cases exhibit complex patterns of spatial heterogeneity in the form of channels, sand lenses, crosscutting faults and/or natural fractures. We have developed a unique stochastic simulation technique in the spectral domain that utilizes polyspectra to reproduce complex spatial patterns of continuity. Some novel approaches to condition these spectral simulations to “hard” data observed along wells will also be presented.
In most reservoir modeling scenarios, the data available to model reservoir heterogeneity is sparse and there is significant prior geologic uncertainty. The practice of model calibration or history matching to update the prior depiction of reservoir heterogeneity is quite popular. In this talk I will present a unique model selection scheme that is a distinct departure from this traditional paradigm of history matching. Instead of using the observed dynamic data to drive an iterative model perturbation scheme, the talk will explore the use of proxy responses to group a suite of prior models into clusters exhibiting similar connectivity characteristics. Then, the cluster exhibiting a flow behavior closest to the observed data is selected using a Bayesian scheme. The resulting selected subset of models permit assessment of residual uncertainty persistent after the model calibration process.
Sanjay Srinivasan is a professor of petroleum and natural gas engineering at the Pennsylvania State University and holds the John and Willie Leone Family chair in Energy and Mineral Engineering. He is also the Department Head for the John and Willie Leone Family Department of Energy and Mineral Engineering.
Srinivasan’s primary research focus is in the area of petroleum reservoir characterization and improved management of reservoir recovery processes. Some of the algorithms and methods that he has pioneered have been applied for early appraisal of ultra-deepwater plays in the Gulf of Mexico and for characterizing natural fracture networks in conventional as well as unconventional reservoirs. He has also partnered with researchers at the UT Institute of Geophysics and the Bureau of Economic Geology to develop novel schemes for integrating seismic data in reservoir models.