Statistical Bayesian Inversion of Azimuthal Resistivity Logging-while-drilling Data
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Learning about the inner earth is one critical objective of earth science. In a modern sense, a better understanding of the geological world supports the basic needs of earth resources. Industrial exploration and production perform drilling practice as one primary process that allows direct contact with the earth formation. Drilling a vertical well straightly down into the earth and extracting hydrocarbon from the reservoir is one frequent exercise in the conventional production. However, the introduction of more cutting-edge technologies makes many attracting applications possible. In the high angle and horizontal activities, the logging-while-drilling application provides controlled directional drilling and navigating through the guidance of real-time formation evaluation. It supports a maximized production from a single well. The vast amount of drilling measurements are obtained through multiple sensing systems and are sent back to the earth surface using a telemetry system. Interpreting the measurements and revealing the true structure of the geological world is never easy. The old experience and traditional mathematical tools seem insufficient to handle the complex, multi-types, and large quantity of the measured data. An effective data-driven solution is necessarily required to guarantee a robust interpretation of complex measurements. Under this background, the dissertation hereby illustrates a class of statistical methods stressing on the geological inverse problems which interpret logging measurements and reconstruct the formation model properties. The methods governed by Bayesian inference are used to reveal the probability of an earth model given the measurements. The solution is returned via a collection of model samples instead of a unique value. It provides a way to assess model uncertainty through the statistics. Several Markov chain Monte Carlo methods are introduced and studied through their implementation of the real-world inverse problems. The methods are generalized to be a data-driven approach that can handle the more advanced tasks that require a reservoir-scale evaluation. It can also deal with the interpretation of model complexity, which lessens the human intervention and guarantees a robust data-driven analyzing scheme.