The main thrust of my research work is solving the combined state and parameter estimation problem for a 3D reservoir model with geological facies using the ensemble Kalman filter (EnKF) technique. The EnKF works very well when the problem is linear and the underlying distribution is Gaussian. One of the major challenges in applying this method to the problem of history matching applied to facies deals with adapting the EnKF to the special case where these two assumptions are strongly violated. Consequently a number of different strategies are investigated to enable the coupling of the EnKF routine to this highly non-linear estimation and history matching problem. More Information
I began my graduate studies in August 2008. My research is focused on the development of techniques to allow the ensemble Kalman filter to be used for data assimilation in very large reservoir models with large amounts of data. I will be investigating improved methods for localization of the Kalman gain and regularization of the covariance.
Rachares Petvipusit (Kurt)
The challenges of determining the type, well schedule, location and the number of wells are among the most important factor in developing a reservoir. An increase in oil and gas consumption requires the best profitable solutions for reservoir management questions. My research started from my curiosity of how to optimize decision methodology whether drilling a new well in optimal location or by applying well conversion optimization. Which wells should be converted? What is the optimal continuous well rate should be controlled? The problems are becoming more challenges if these issues are incorporated together. At the beginning phase of research, I will be investigating the efficient and robust method for these issues.
Well Placement Optimization
Starting in January of 2010, I am working as a research assistant for this project. Our primary objective is to develop methods for inversion of data collected by nanosensors to estimate reservoir rock and fluid properties. The focus of the inversion is on estimation of the three-dimensional distribution of fluids and rock properties as well as the resolution of the estimate and a quantification of the uncertainty. This will aid in understanding which aspects of the sensor transport and measurement function are critical for high resolution illumination of static 3-D properties and dynamic flow paths.
Research interest lies in better quantificaiton of the uncertainties existing in initial ensemble of parameters, model structure, and data when applying the ensemble Kalman filter technique for history matching. Quantifying uncertainties in a proper way is very important for achieving good history matching results and future performance prediciton, however, most of the time, uncertainties are underestimated or estimated with bias. More Information
During the analysis step of ensemble Kalman filter (EnKF) method, spurious correlations often lead to unrealistic changes in porosity and permeability fields. Currently, I am working on a new approach to minimize the effect of spurious correlations by using localized basis vectors for updating. The compositional module, Eclipse 300 from the Schlumberger’s reservoir simulation suite is being used for carrying out the forward model evaluations. It can be shown that, when the entire reservoir model is subdivided into moderate size local regions, the local reservoir dynamics is often low dimensional. The usefulness of this approach has been demonstrated on a synthetic 2-dimensional (vertical cross-section) compositional reservoir model with a total of 12 components (11 components from oil phase and 1 from water). Thus, each grid block in the reservoir model has 15 state variables (porosity, logarithm of permeability, pressure and the 12 components mentioned above). The newly proposed technique gave slightly better history matching results compared to the standard EnKF for a synthetic problem. More Information
My current research mainly focuses on the production optimization for optimizing the reservoir performance. Even though the numbers of applications for the optimization methods are rapidly increasing, there are a number of challenging issues that need to be worked out in order to make the applications robust and more efficient. Some of these issues involve moving horizon approach for well controls, different types of regularization or parameterization to obtain piecewise constant controls, handling of inequality constraints during optimization, reducing the overall cost of the optimization process, incorporating discrete events such as workovers or drilling, as well as the cost of information in the optimization algorithm. I am going to address some of these issues during the early phase of my research.
For the past few years I have primarily worked on the application of the Ensemble Kalman Filter to the problem of production data assimilation or history matching in petroleum reservoirs. I am particularly interested in developing efficient methods for dealing with nongaussian probability distribution in the static reservoir model parameters (e.g. permeability and porosity ) and in the dynamic variables (e.g. saturations and concentrations). Problems of particular interest include iterative filters, reparameterization, facies models, localization, clustering, and closed-loop optimization.