Subsurface data and AI in Equinor

Binge watch online from 3,000 meters below ground? You can if you are a subsurface specialist turned data scientist.

Are you able to watch and follow the action in two movies at the same time? Probably not for long. What about 10,000 Netflix movies per second? Of course not. Too much for our natural intelligence. On our fields we generate such amounts of data from our operations every day, all day long. Now engineers and geoscientists are turning to data science to get more out of the subsurface.

Equinor has placed fiber optic cables, the very same type of cable that delivers the Internet to your home, along all the wells at the giant Johan Castberg field. While there’s little need for Netflix in the middle of a reservoir, there is a need for data and that is why we are using these cables. With slight modifications, the fiber now works as microphones and temperature sensors which create enormous amounts of data.

Taber Hersum, a geologist turned data scientist, can “binge” watch this and other huge data sets that are being generated with new technology and sensors in our operations, and make sense of it.

“My eyes have been opened to data-driven approaches. It is so intriguing because we can make insight into complex relationships which might otherwise be impossible without data science”, Taber says.

When all this data is gathered in one place, 10,000 “movies” can be made into 1 that our engineers can watch. Process capabilities of the subsurface is translated into information with the help of machine learning. What used to take 2-3 weeks and involved flying hard disks to shore with helicopter for analysis, can now be done in seconds and livestreamed. And when the data is live, it opens for automation. There is no need to watch all the data as it comes up from the deep below. The fiber team is developing machine learning algorithms to make it easier for the engineers to view the right data.

“The implementation of a real-time data streaming pipeline for fiber optic data has enabled much more efficient interpretation of the vast amounts of data we have. Many observations can readily be interpreted by subject matter experts, and we reuse those interpretations as training data to build machine learning models. This is a critical step towards automated interpretation and will be increasingly important as the data continues to grow,” Taber explains.

“If we can better represent the complexity of well planning including the available slots, the targets, and all the constraints, then this means value creation and efficiency with our precious time. For me, in AIM, that means working across subsurface and drilling and well challenges – I have to understand each other's data and knowledge in order to get the cheapest and least risky well paths”, Roy explains.

Part 1.
In traditional seismic mapping, manual interpretation is done at regular intervals (one ’stripe’ is one interpreted line, 400 meter apart) and a map is generated. The quality of the map depends on the density of interpretations. This area was manually mapped in three hours. This area is 700 square kilometres.

Part 2.
With machine learning (image recognition), our algorithms only need a few seismic lines to map the same area. This interpretation takes 5-10 minutes. The algorithms learn from the analysis done by our experts

Part 3.
This shows the quality of the map from traditional manual mapping techniques. The geologist has interpreted every 32nd line. The next picture shows the result after using machine learning on the same task.

Part 4.
The algorithm interprets ALL lines, not just every 32nd line. This gives a much more accurate and detailed map. The compute time to achieve this result is less than one hour. Machine learning helps the geologist to reach analytical results faster, freeing up time for more in-depth analysis.

The huge data potential

Before well planning and drilling, we need seismic data. In fact, we already have 50 petabytes of it from around the world. That is more than double the amount of digital content held by the US Library of Congress, the world’s largest library. We are on a journey to standardise our data and make it easily available through the cloud. When historical data can be combined with newer data, geology from different geographies can be analysed and compared to look for patterns and potential, and when a geologist can use software to analyse vast amounts of data that would take years to go through manually, it changes the game.

Renaud Laurain, a geophysicist who has also upskilled himself into data science has been working in several projects where the aim is to make seismic analysis easier:

“Moving from seismic data acquisition to data analytics has given me a completely different point of view of the importance of data in the modern world. Here is one example: In the past our work processes were focused on reducing the amount of data so that a human being can handle it. Data analytics and machine learning change that entire objective – shifting to focus on how to capitalize on the whole data amount,” he explains.