By: Tom Smith
Most people associate neural networks, big data and big number crunching as parts of a single paradigm for access to web information. Articulate a query and wait for a list of answers. But in the oil and gas exploration and reservoir replacement business – particularly at this time – we “must” place neural networks and big data tools in the hands of seismic interpreters.
Seismic interpreters are accustomed to working on interactive workstations, not using web-based queries. We suggest this “must” not because seismic interpreters are a narrow-minded bunch who are unwilling to work with newer web-styled tools. No – there are strong technical reasons which will allow us to dig deeper into our seismic data. The key is not the platform, be it an interactive workstation or the web. The real reason is that we simply do not and cannot understand much about multi-attribute seismic data with conventional seismic interpretation methods – we don’t know the semantics of the words in these data. Instead, we must let neural networks fly through the seismic data unattended to discover whatever properties they might discover. In other words, multi-attribute seismic data ain’t English. We don’t know the language yet. Today, we are starting to use web-style tools to find things in seismic data in just the same way they search through web pages looking for word patterns.
Someday integration of machine learning and seismic interpretation will evolve into a practice no longer considered a novel but a mundane operation. Then, articulation of queries like, “In this 3D survey, are there any geobodies with statistical properties of Miocene fluvial channel sands?” will be practical. In fact, a large multinational oil company with a world-wide petabyte seismic library might ask a query, “In the eastern Gulf of Mexico, do we have any 3D surveys with geobodies that might be Miocene fluvial channel sands?”
Queries like this will take lots of number crunching – recognizing and indexing geobodies in 3D surveys and all this will be done long before queries are made – but the web paradigm of today applies directly to the seismic interpretation of tomorrow. In terms of technology crossover, it’s easier than falling off a log. Just time, people and money…We happen to be a little ahead of the pack.
We have been told that at a presentation at a major oil company a particular slide drew much attention. On that slide was drawn two curves – curves that must be part of some internal assessment. One curve was total volume of annual digital seismic data acquired and processed and the second curve was the volume of these data which has been actually inspected. The two curves diverged and they were concerned. They realize that the solution to this kind of problem is addressed by Big Data and therefore, presentations have been made by HP, IBM and now us, Geophysical Insights.
The beauty of web information access today is that it is non-confrontational. For a well-posed question, there are often many potential answers. These are ranked in decreasing order of importance and suspected relevance. We’ll be able to do the same with seismic data.
So what we are building in Paradise® today is a workbench for practitioners. The process of neural network analysis is not run once to deliver some magic answer. It is run many times with each run under the careful eye of an interpreter. And our Geophysical Insights geoscientists say that the most challenging part of learning to use this new technology is to appreciate the results. The more seasoned the interpreter => the higher the likelihood that the results of machine learning will have some meaning in the eyes of the interpreter. These new tools are part of a new kind of seismic interpretation where fresh eyes bring new insight. Sometimes significant results are not obvious but often the obvious oil & gas prospects have been evaluated. Subtle but significant results are OK too but if and only if the drill proves out a promising prediction.