Log-facies classification using expectation-maximization
Xiaozheng Lang, Mita Sengupta, Chicheng Xu, Tianqi Deng, Darío Graña · 2019
The goal of log-facies classification is to predict the vertical profile of lithological facies at the well location. Different facies generally show different elastic and petrophysical properties in well log measurements. Deterministic or probabilistic methods for cluster analysis are generally applied to classify the samples in the well log in a given number of facies. For example, Bayesian classification, or Naïve Bayes method, is often used to classify well logs into facies. However, the majority of these methods does not account for the vertical continuity in the facies sequence. In traditional facies classification methods, each sample is classified independently from the position in the profile; in other words, the facies assigned to a given sample does not depend on the facies assigned to the sample above nor the sample below. In this work, we present a method, namely Expectation-Maximization, based on Markov chains to describe the spatial continuity in the facies profile. The Expectation-Maximization method allows the classification of facies at each location of the well logs, based on a set of measured properties, i.e. the well log data, and the facies classification in the upper layer, i.e. the facies sample above the given location. The proposed method is a probabilistic classification algorithm and geological information can be included in the prior model. In the proposed implementation, the parameters of the Markov chain are assumed to be unknown (hidden Markov model) and are estimated simultaneously with the classification. The methodology was validated on a set of well logs for a CO2 sequestration study and compared to the Naïve Bayes method. Presentation Date: Monday, September 16, 2019 Session Start Time: 1:50 PM Presentation Start Time: 1:50 PM Location: 217A Presentation Type: Oral