In unconventional resource plays, constructing a sound geological model that ties various well information is imperative for properly extracting and integrating well and seismic information and for predictive and prescriptive analytic workflows. Unlike conventional plays, unconventional plays that span basins have potentially tens of thousands of wells. Constructing geological models to include all wells and then updating them as additional ones become available can be a daunting task.
When constructing large cross sections, regional stratigraphic patterns are easily discernible visually. Converting these geologic events and spatial patterns to digital information using the power of the computer and new machine learning techniques is becoming more important than ever as geoscientists attempt to "keep up" with all this information. This paper will cover a modern technology toward that end.
Previous attempts have been made to pick geologic well tops automatically using expert systems (Olea et al.), neural networks (Luthi et al.), and dynamic programming (Lineman et al., Inazaki, Zoraster et al., Fang et al.). While these previous efforts have been helpful in defining the problems and building blocks to solve well-log correlation automatically, they have clearly been much less successful than has been observed in seismic picking algorithms that started in the 1980's. This is mainly owing to the nature of seismic data. Seismic traces are band-limited, closely spaced (on the order of meters) with neighboring traces almost identical to each other, and are consistent with the same start and ending times, sample rates, and vertical representation. These traits make correlating neighboring peaks, troughs and zero-crossings reasonably easy as compared to well logs, which are more widely spaced (on the order of hundreds to thousands of meters), have inconsistent depth ranges with possible gaps, and may be from highly non-vertical well bores.
As more oil companies transition from exploration to resource recovery optimization and the number of new wells in well-known basins dramatically increases, geologic cross sections across these basins begin to take on more of a seismic look, as shown in Figures 1 and 2 below. When logs are hung on stratigraphic datums, as Figure 2 shows, geologic intervals are readily evident across many tens, if not hundreds or thousands of wells. Not only is the lateral consistency of strong events evident, such as the Codell in this case, but patterns of finer detail in the sequence stratigraphy (flooding surfaces, onlap, thickening and thinning from changing accommodation and sediment supply) become more visually apparent. Further refined picking of associated events is warranted but could prove tedious and time consuming if done manually.
