Automatic Well Log Correlation
Abstract
Unconventional plays can have thousands of wells. These wells are an important source of information for creating a subsurface stratigraphic model. To understand the subsurface data from these well logs, the geologist is tasked with quickly correlating wells with multiple well-log curves; however, it might take the geologist several days to manually correlate the well logs and produce several surface picks. This paper presents a newly developed automated well-log correlation algorithm to correlate well logs within the same depositional system using an enhanced dynamic time-warping algorithm. This method uses computer vision techniques to recognize patterns, in addition to log curve values. Unlike previous dynamic time warping techniques, the algorithm does not need a constraining surface pick to perform the correlation; it can leverage all of the different well-log curves simultaneously and generate multiple surface picks. In addition, well logs do not need to be supplied to the algorithm in a cross-section along a depositional dip or in a geologically significant order, but can be input in a random order. The algorithm outputs as many surfaces picks as requested, and the picks can be used to create multiple surfaces. Well-log data are notoriously noisy; therefore, the automated surface picks might not be perfect. Thus, the algorithm includes a system to obtain user input and update its correlated surface picks accordingly.
AAPG Datapages/Search and Discovery Article #90291 ©2017 AAPG Annual Convention and Exhibition, Houston, Texas, April 2-5, 2017