Machine learning–based method for automated well-log processing and interpretation

Abstract

Traditional well log processing and interpretation workflows are subjective, time-consuming, and can be inconsistent. To improve the objectivity, efficiency, and consistency of these workflows, we proposed a machine learning-based method that automates some key workflows, such as zonation assignment, outlying data detection, and formation property interpretation. We developed the cross-entropy clustering-Gaussian mixture model-hidden Markov model workflow that discovers locally stationary clusters equivalent to stratigraphic zones, and then propagates zonation information from training wells to other wells. The same workflow also flags potential outlying data. For interpretation, we developed the cluster-based (i.e., zone-based) predictive modeling approach that mimics the traditional interpretation workflow. The results show that our method can discover clusters (or zones) in training data, and then propagate the zonation to other wells. Presentation Date: Wednesday, October 17, 2018 Start Time: 8:30:00 AM Location: 204B (Anaheim Convention Center) Presentation Type: Oral